Sellforte Solution Research
Research notes: Top MCP Servers for Marketing Mix Modeling and Incrementality Testing
Evaluation date: 23 September 2026
Read the article using this research: Top MCP Servers for Marketing Mix Modeling and Incrementality Testing.
This page contains the original Claude and ChatGPT evaluations behind Sellforte’s research into MCP capabilities for marketing mix modeling and incrementality testing of different vendors. The evaluations cover eight vendors and 48 criteria per vendor, using public information reviewed on 23 September 2026. Each vendor’s Claude evaluation appears first, followed by its ChatGPT evaluation. The tables reproduce the individual evaluation sheets, including scores, rationale, source URLs and source types.
Sellforte designed and published the research and is one of the vendors assessed. Claude and ChatGPT assigned the scores and wrote the evaluation notes. These are assessments of public evidence, not hands-on product tests. A low or zero score may reflect limited public documentation rather than an absent capability.
Evaluation criteria
| Category ID | Category | Criteria ID | Criteria | Description |
|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | MCP reports actual online sales for a specified period. |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | MCP reports actual physical-store sales for a specified period, using offline store sales data retrieved from customer's data warehouse. |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | MCP reports digital media data across Meta, Google, TikTok, and other major paid platforms. |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | MCP reports spend and media metric data for offline media, covering at least TV, out-of-home, radio, print. |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | MCP reports supported business outcomes such as contribution margin, orders or new customers. |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | MCP applies explicit date, brand, market, product and campaign filters and returns results at the requested supported granularity. |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | MCP reports true incremental impact, not just last-click or platform-reported ROAS, for each digital channel. |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | MCP reports incremental ROAS and revenue for offline channels such as TV, OOH, and radio. |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | MCP surfaces how promotions and pricing changes contributed to sales, not just paid media. |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | MCP provides daily measurement of incremental ROAS based on MMM, not just weekly, monthly or quarterly model refreshes. |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | MCP exposes modeled contributions from media, baseline and supported non-media drivers to explain changes in business outcomes. |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | Invokes the provider’s optimization engine through MCP to calculate channel budgets for a specified objective and period. |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | Invokes the provider’s forecasting engine through MCP to estimate outcomes for a specified budget allocation. |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | Retrieves marginal returns and spend-response data for supported channels through MCP. |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | Calculates the modeled impact of a specified change to channel spend through MCP. |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | Passes user-defined budget limits, channel restrictions and planning dates to the optimization engine and exposes the resulting constraints. |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | Returns comparable spend and outcome estimates for alternative scenarios and an explicit baseline plan through MCP. |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | Calculates a budget plan for a supported target such as incremental profit, customer acquisition or a specified outcome level. |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | Retrieves previously saved scenarios through MCP with their identifiers, inputs and outputs. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | MCP reports incremental revenue and ROAS at the individual campaign and ad set level, not just at the channel level. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | Returns comparable incremental, last-click and ad-platform returns for supported campaigns and ad sets. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | Retrieves modeled marginal returns for individual campaigns and ad sets through MCP. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | Returns daily budget recommendations through MCP. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | Returns bidding targets (e.g., Target ROAS) through MCP. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | MCP can execute daily spend/budget changes directly on major ad platforms via API. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | MCP can execute bidding parameter changes (e.g. Target ROAS) directly on major ad platforms via API. |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | After a bidding or budget change is applied, MCP can provide the actual impact using a pre/post comparison for bidding change at the campaign & ad set level. |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | Retrieves completed geographic incrementality-test results with the tested intervention and outcome identified. |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | Retrieves incrementality results for randomized or controlled owned-media tests such as email or leaflet experiments. |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | Retrieves incrementality results from advertising-platform lift studies such as Meta Conversion Lift. |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | Uses available measurement results and uncertainty to recommend specific hypotheses, channels or markets for testing. |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | Produces a test design specifying treatment and control, power or minimum detectable effect, duration and required spend. |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | Provides documented MCP connection and authentication instructions for Claude and ChatGPT |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | Returns server-provided tables or chart artifacts that a documented compatible MCP client can display. |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | Makes available discoverable MCP tool definitions and documentation for interpreting the returned data. |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | Provides complete structured results or downloadable data files for use in spreadsheets, reports and downstream tools. |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | Documents repeatable MCP invocation from an external agent runtime for recurring analysis workflows. |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | MCP-provided recommendations are grounded in a Bayesian Marketing Mix Model, not last-click attribution or descriptive analytics. |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | The MMM is calibrated against real incrementality test results, priors and posteriors informed by causal experiments. |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | The tool reports model validation metrics (R2, MAPE, posterior predictive checks, holdout performance) so users can assess model quality. |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | Customers can inspect and configure model priors and other key parameters in a self-serve UI, not just accept the model as a black box. |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | Allows authorized agents to propose or apply model-calibration changes through MCP with validation and version history. |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | Proven track record with large, sophisticated advertisers, not just mid-market or DTC brands. |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | Independently verified security posture, a baseline requirement for enterprise IT procurement. |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | Customers can choose where their data is stored, critical for GDPR compliance in Europe. |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | Deployment flexibility to match the customer's existing cloud infrastructure. |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | Enterprise authentication via SSO, required by most large-company IT policies. |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | A real, promtable demo of the MCP capabilities is publicly accessible without requiring a sales. |
Evaluation instructions
| Instruction ID | Instruction |
|---|---|
| 1 | You are an independent evaluator of MCP servers for Marketing Mix Modeling and Incrementality testing |
| 2 | Use these evaluation criteria for categories 1-5: 1.00: Strong public evidence that the capability exists in the MCP 0.75: Partial public evidence that the capability exists in the MCP 0.50: No evidence found that the capability is available via the MCP, but there is strong evidence that the capability is available in the broader platform 0.25: No evidence found that the capability is available via the MCP, but there is partial evidence that the capability is available in the broader platform 0.00 : No evidence found that the capability exists in the MCP or in the broader platform |
| 2 | Use these evaluation criteria for categories 6-8: 1: Strong evidence that the platform supports the capability 0.5: Partial evidence that the platform supports the capability 0: No evidence found that the platform supports the capability |
| 3 | In the evaluation, only use information available at the company website, and in the domain where technical documentation is located (if separately hosted). |
| 4 | Prioritize type of source materials in this order 1. Technical documentation (such as support center) 2. Product page 3. Marketing collateral (such as product launch blog posts) |
| 5 | When interpreting terminology, you can assume that - ROI is the same thing as incremental ROAS or iROAS - Marginal ROI is the same thing as Marginal Incremental ROAS or miROAS - Diminishing return curves are the same thing as response curves |
| 6 | For the purposes of this evaluation, feature available via API also qualify with same scores as features via MCP. |
| 7 | As an output, provide your evaluation in an excel file, using these columns: - Category ID - Category - Criteria ID - Criteria - Criteria Score - Two-sentence rationale for the score. If you found no evidence, comment that you did not find evidence, instead of claiming that the capability does not exist - URL to source - Type of source (Technical doc, Product page, Marketing collateral) |
Fospha
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The MCP Performance Summary tool returns revenue, conversions and ROAS for a requested period, and Fospha revenue is reconciled to the eCommerce platform so it reflects all actual online orders. Amazon sales are excluded by default but can be included on request. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.25 | We did not find evidence that the MCP or platform ingests a brand's own physical-store sales from a customer data warehouse. The only related evidence is a Sephora retailer connector described as covering Sephora US online and in-store sales. | https://www.fospha.com/platform/integrations | Product page |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The MCP Channel Breakdown, Spend Allocation and Campaign Drill-Down tools return spend and performance by channel and source, such as Paid Social Meta and Paid Search Google. The platform has 60+ connectors including Meta, Google Ads, TikTok, Snapchat and Pinterest, with impressions, clicks, CTR and CPM defined in the metrics glossary. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.25 | We did not find evidence that the MCP reports TV, out-of-home, radio or print spend and metrics. The platform lists connected TV connectors (e.g. MNTN, Tatari) and an Enterprise-only offline spend calibration, which is partial evidence at best. | https://www.fospha.com/platform/integrations | Product page |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The MCP Performance Summary returns conversions, CPP, CAC and AOV alongside revenue, and the glossary defines new-customer conversions and new-customer revenue. We found no evidence of contribution margin or profit as a reported outcome. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | The MCP supports explicit date ranges, market and channel breakdowns, campaign filters (name, type, strategy, objective) and brand selection for users with several brands, at daily, weekly or monthly granularity. We found no native product dimension; product splits rely on rule-based custom categories built from campaign names. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The MCP skill library documents that the Beam data tool returns incremental revenue and incremental ROAS per channel, and Fospha ROAS is produced by its daily model rather than last-click. The MCP Attribution Comparison tool also contrasts Fospha with last-click and ad-platform figures by channel. | https://help-center.fospha.com/help-center/ask-fospha-ai-skill-library | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.25 | We did not find evidence that the MCP reports incremental ROAS or revenue for TV, OOH or radio. The platform mentions offline calibration and offline spend inputs to the model, which is only partial evidence of offline channel measurement. | https://www.fospha.com/platform/core | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0 | We did not find evidence that the MCP or the platform reports promotion- or pricing-driven revenue as a separate contribution. Promotions appear only as a model input, as sale-period flags in Beam and as annotations for explaining metric movements. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 1 | Fospha's impression-level Daily MMM is retrained every day and core modelled outputs are delivered every 24 hours. The MCP guide states the data is refreshed daily, although Beam saturation curves update every few days. | https://help-center.fospha.com/help-center/how-does-fospha-validate-its-model | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.25 | We did not find evidence that the MCP exposes a base versus media versus non-media decomposition. The platform reconciles 100% of sales across channels including organic and direct, and only the Enterprise beta Brand Impact module models a seasonal baseline. | https://help-center.fospha.com/help-center/reconciliation | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | We did not find evidence of an optimizer tool in the MCP or API. The platform's Budget Planner (beta) returns recommended spend per channel for a chosen market, KPI, budget and timeframe. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.75 | The MCP Beam Spend Forecasting tool forecasts ROAS, CPP or CAC for a target daily spend on a specific channel. We found no evidence of forecasting a complete multi-channel media plan through the MCP. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 1 | The MCP Beam Spend Optimisation tool returns channel headroom, scaling status, saturation points and saturation curves. The API's incremental forecasting summary also returns saturation point and headroom per channel segment. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 1 | The MCP Beam Spend Forecasting tool models what-if changes to daily spend on a channel, with the documented example of raising Meta daily spend to a set amount and returning expected ROAS. Beam curves in the platform show low, most-likely and high outcomes for a chosen spend. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.5 | We did not find evidence of constraint-aware planning through the MCP. In the platform, Budget Planner takes a total budget, channel inclusion and timeframe and applies fixed per-cycle shift guardrails, but user-set per-channel minimums and maximums are not documented. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.25 | We did not find evidence of scenario comparison in the MCP; its Compare Periods tool compares historical periods, not plans. Budget Planner shows current versus recommended spend, which is partial evidence of a baseline comparison. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.5 | We did not find evidence that the MCP calculates a budget plan against a target, although its Targets tool returns ROAS, CAC or CPP targets. Budget Planner optimizes toward a chosen KPI such as ROAS, CAC or new-customer ROAS; we found no profit objective. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0.25 | We did not find evidence that the MCP or API retrieves saved plans. The Budget Planner doc advises naming plans so you can recognize them later and includes a plan library screenshot, which is partial evidence of saved plans in the platform. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The MCP Campaign Drill-Down tool returns campaign, ad set and ad-level performance including Fospha ROAS from its daily model. Fospha labels this modelled attribution rather than explicitly incremental ROAS, which it documents only at channel level. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 1 | The Fospha API exposes Fospha ROAS, ad-platform ROAS and last-click ROAS metrics that can be grouped by ad set and ad. The MCP Attribution Comparison tool compares the same three models, and the Optimization Dashboard shows them side by side per campaign and ad set. | https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-metric | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.25 | We did not find evidence of marginal ROAS for individual campaigns or ad sets through the MCP. Beam saturation curves exist per channel segment (e.g. Meta prospecting), which is partial evidence below campaign level. | https://help-center.fospha.com/help-center/beam-incremental-forecasting | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | We did not find evidence of per-campaign daily budget recommendations through the MCP. In the platform, Ask Fospha AI Budget Actions (closed beta, TikTok only) proposes daily budget changes for campaigns with reasoning and a forecast. | https://help-center.fospha.com/help-center/fospha-ai- | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0 | We did not find evidence that the MCP or platform recommends bid targets such as Target ROAS. The Budget Actions doc states that it does not re-bid campaigns. | https://help-center.fospha.com/help-center/fospha-ai- | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.25 | We did not find evidence that the MCP pushes budget changes to ad platforms. Budget Actions in the platform can apply approved daily budget changes on TikTok only (closed beta), and the Prism integration lets Smartly reallocate budgets on Meta and other platforms. | https://help-center.fospha.com/help-center/fospha-ai- | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | We did not find evidence that the MCP or platform pushes bidding changes such as Target ROAS to ad platforms. Budget Actions explicitly excludes re-bidding campaigns. | https://help-center.fospha.com/help-center/fospha-ai- | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | We did not find evidence of pre/post impact analysis through the MCP. Budget Actions in the platform tracks outcomes against a steady-state baseline with drift alerts and an audit log, but only in closed beta for TikTok budgets. | https://help-center.fospha.com/help-center/fospha-ai- | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.5 | We did not find evidence that the MCP or API retrieves geo test results. Fospha offers geo-lift tests designed and run by its marketing science team, with results and recommendations fed back into the model. | https://www.fospha.com/platform/incrementality | Product page |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0 | We did not find evidence of owned-media A/B test results (e.g. email or leaflet experiments) in the MCP or the platform. Fospha's documented testing offer centers on geo-lift tests run by its science team. | https://www.fospha.com/platform/incrementality | Product page |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.25 | We did not find evidence that the MCP retrieves Meta Conversion Lift results. A case study describes ingesting live conversion lift results to calibrate the model, and pricing lists Enterprise lift calibration. | https://www.fospha.com/case-studies/carparts | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | We did not find evidence that the MCP prioritizes experiments. The platform's science team helps identify which questions to test, and the Beam Prediction Confidence doc suggests using confidence to prioritize channels to investigate. | https://help-center.fospha.com/help-center/the-prediction-confidence-score | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.25 | We did not find evidence that the MCP designs incrementality tests. The incrementality product page mentions a test design engine that finds matching regions, but we found no documented power, MDE or required-spend calculation. | https://www.fospha.com/platform/incrementality | Product page |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | The help center gives step-by-step setup for Claude (web, desktop, Team and Enterprise via the connector directory) and ChatGPT (Free to Enterprise via Developer mode), with regional server URLs and OAuth login. Product-page FAQs are less current and still describe ChatGPT as coming next. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | We found no evidence of server-provided tables or chart artifacts from the MCP. The skill library instructs the AI client to render results as Markdown in chat. | https://help-center.fospha.com/help-center/ask-fospha-ai-skill-library | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | The MCP includes a Fospha Intro tool that returns a guide to all tools, parameters and recommended workflows, and the skill library names the underlying tools. A public metrics glossary explains how to interpret returned metrics. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | Fospha offers a GraphQL export API with OAuth2 client credentials for loading performance data into BI tools and warehouses such as Snowflake and BigQuery. The platform also supports scheduled exports to email, Google Sheets, SFTP and S3. | https://help-center.fospha.com/help-center/fospha-api-programmatic-access-to-your-performance-data | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | Marketing material positions Fospha for Agents and the MCP for automated workflows and reporting pipelines, and the API is pitched for AI agents. We found no technical guide for invoking the MCP from an external agent runtime on a schedule. | https://www.fospha.com/blog/ask-fospha-ai-is-your-marketing-strategist-inside-the-fospha-platform | Marketing collateral |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | The technical docs describe an ensemble model whose impression-level Daily MMM uses XGBoost with Shapley values, with Bayesian methods documented only for Beam saturation curves and Brand Impact. Marketing material refers to a Bayesian MMM, so the evidence for a Bayesian MMM backbone is partial. | https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 0.5 | Product pages state that incrementality test results feed back into the model as calibration inputs, and Enterprise pricing lists lift calibration of saturation curves (partly beta). We found no technical documentation of how tests inform priors. | https://www.fospha.com/platform/incrementality | Product page |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 0.5 | The docs describe daily retraining, testing on unseen data, back-testing and RMSE monitoring, with NRMSE available on request and a Beam Prediction Confidence score for selected customers. We found no evidence of MAPE, R2 or posterior predictive checks reported to users. | https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0 | We found no evidence that priors or model configuration can be inspected or edited in a self-serve UI. Documented user settings cover targets, custom metrics, channel groupings, Beam sale-period flags and annotations. | https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We found no evidence that model calibration can be proposed or applied through the MCP or API. MCP writes are limited to logging and updating annotations, and the API is read-only. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Fospha shows logos of large brands such as Lululemon, Dyson, Urban Outfitters and Samsonite, and publishes case studies for Galeries Lafayette and Callaway Golf. Most named references (e.g. Gymshark, Huel, River Island) are below $1B revenue, so we found fewer than 10 $1B+ references. | https://www.fospha.com/case-studies | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Fospha's trust center lists ISO 27001:2022 as certified, along with GDPR and CPRA compliance. SOC 2 is shown as coming soon. | https://trust.fospha.com | Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0.5 | The trust center states that UK, EU and international systems are hosted in AWS UK regions and US systems in AWS US regions, with separate UK and US app and MCP endpoints. The non-US option is the UK rather than an EU region, and we found no evidence that customers choose the region themselves. | https://trust.fospha.com | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We found no evidence of a choice between AWS, GCP and Azure. The trust center and application privacy policy describe hosting on AWS. | https://trust.fospha.com | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0.5 | The MCP guide says users log in with Fospha credentials via SSO or email and password. We found no documentation of SAML, OIDC or identity providers such as Okta or Microsoft Entra. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We found no evidence of a public, promptable MCP demo or trial; the MCP requires a fully onboarded Fospha account. The MCP product page routes users to a demo booking or their account manager, and pricing lists MCP under the Enterprise tier. | https://www.fospha.com/platform/fospha-ai-automation | Product page |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The API returns dated revenue totals, and Fospha documents reconciliation to actual ecommerce sales. This supports online sales reporting for a selected period. | https://api-docs.uk.fospha.com/api/marketing-performance/operations/queries/marketing-aggregate https://help-center.fospha.com/help-center/fospha-vs.-ad-platform-reporting | 1. Technical doc 2. Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0 | Published sales-channel coverage focuses on ecommerce, marketplaces and apps. I found no evidence of physical-store sales reporting from a customer warehouse through MCP, API or the broader platform. | https://www.fospha.com/pricing https://www.fospha.com/platform/integrations | 1. Product page 2. Product page |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | API examples return cost, impressions and clicks by source and date. Fospha documents integrations with Meta, Google, TikTok and other paid platforms. | https://api-docs.uk.fospha.com/guides/examples https://www.fospha.com/platform/integrations | 1. Technical doc 2. Product page |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.25 | Enterprise pricing lists offline-spend calibration, providing partial platform evidence for offline media inputs. I found no reporting evidence covering TV, out-of-home, radio and print together through MCP or API. | https://www.fospha.com/pricing | 1. Product page |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The API exposes conversions, new-customer conversions and subscription conversions. These provide explicit business outcomes beyond revenue. | https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-metric | 1. Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 1 | The API documents account selection, date ranges, market and campaign filters, plus grouping dimensions. Custom Categories adds a documented API reporting axis for configured product divisions and business units. | https://api-docs.uk.fospha.com/api/marketing-performance/operations/queries/marketing-aggregate https://api-docs.uk.fospha.com/api/marketing-performance/types/inputs/marketing-filter https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-dimension https://help-center.fospha.com/help-center/how-to-create-a-custom-category- | 1. Technical doc 2. Technical doc 3. Technical doc 4. Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | Fospha’s published MCP workflow retrieves incremental revenue and incremental ROAS per channel using get_beam_data. It explicitly separates these from Fospha-attributed revenue. | https://help-center.fospha.com/help-center/ask-fospha-ai-skill-library | 1. Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0 | The product describes offline calibration alongside digital measurement. I found no evidence of reported incremental revenue and ROAS for individual offline channels through MCP, API or platform outputs. | https://www.fospha.com/platform/core https://www.fospha.com/solution | 1. Product page 2. Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.25 | Core incorporates promotions, while annotations identify sale periods and support forecasting exclusions. I found no quantified promotion-driven revenue output through MCP or API, so the evidence is only partial at platform level. | https://www.fospha.com/platform/core https://help-center.fospha.com/help-center/how-to-use-annotations-in-fospha | 1. Product page 2. Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.75 | Core refreshes daily and its measurements are accessible programmatically, supporting part of this criterion. Beam’s technical guide says its model updates every few days, leaving strictly daily incremental-ROAS refresh unconfirmed. | https://www.fospha.com/platform/core https://help-center.fospha.com/help-center/beam-incremental-forecasting https://help-center.fospha.com/help-center/fospha-api-programmatic-access-to-your-performance-data | 1. Product page 2. Technical doc 3. Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.25 | Model Spotlight explains channel attribution changes through a model waterfall. I found no explicit decomposition of base demand, media and non-media contributions exposed through MCP or API. | https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | Budget Planner calculates channel allocations for a selected KPI, budget and timeframe, with beta availability disclosed. I found no documented MCP or API operation invoking that portfolio optimizer. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.75 | MCP forecasts outcomes for a specified channel spend. I found no documented invocation accepting an entire user-specified media allocation. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | MCP exposes saturation curves and channel headroom. I found no explicit marginal-ROI output, so the combined criterion is partially evidenced. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 1 | MCP accepts a target daily channel spend. Its forecasting tool returns modeled ROAS, CPP or CAC for that change. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.5 | Budget Planner accepts a total budget, channel selection and timeframe, and applies allocation guardrails. I found no MCP or API optimizer invocation passing these constraints. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.5 | Budget Planner compares current and recommended channel spend with predicted KPI impact, and supports reruns at different budgets. I found no MCP or API operation returning these portfolio comparisons. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | 1. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.25 | Budget Planner optimizes a selected KPI, and Beam evaluates forecasts against CAC or ROAS targets. I found no documented MCP or API solver that calculates a plan to achieve a specified business-outcome level. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner https://help-center.fospha.com/help-center/beam-incremental-forecasting | 1. Technical doc 2. Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0.25 | Budget Planner documents named plans and depicts a plan library, suggesting partial platform support for saved scenarios. I found no MCP or API retrieval of saved plan identifiers, assumptions and outputs. | https://help-center.fospha.com/help-center/how-to-create-a-budget-plan-with-budget-planner | 1. Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The API exposes Fospha revenue and ROAS at campaign and ad-set granularity. Its methodology allocates aggregate MMM results using finer platform signals, leaving independently modeled incremental effects at each requested level only partially evidenced. | https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-metric https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-dimension https://www.fospha.com/blog/can-a-media-mix-model-provide-reliable-guidance-at-the-ad-or-creative-level | 1. Technical doc 2. Technical doc 3. Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | The API supports campaign and ad-set grouping with Fospha, last-click and ad-platform ROAS. The Fospha view is hybrid attribution, and I found no equivalent three-way comparison using distinct incremental ROAS at both levels. | https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-metric https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-dimension https://www.fospha.com/blog/can-a-media-mix-model-provide-reliable-guidance-at-the-ad-or-creative-level | 1. Technical doc 2. Technical doc 3. Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0 | Beam’s API documents channel-segment saturation and headroom. I found no campaign- or ad-set-level marginal-ROI output through MCP, API or the broader platform. | https://api-docs.uk.fospha.com/api/incremental-forecasting/types/objects/incremental-forecasting-summary-point | 1. Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | Prism uses Fospha signals in Smartly to allocate daily budgets at campaign or ad-set level. I found no exposed MCP or Fospha API operation returning those optimal budgets. | https://help-center.fospha.com/help-center/prism-fospha-smartly-how-to-set-up-a-smartly-pba-test-using-fospha-data | 1. Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | Fospha describes measurement-informed bidding in its Auction integration. I found no documented numerical bid-target recommendation for each campaign or ad set through MCP or API, leaving only partial platform evidence. | https://www.fospha.com/platform/fospha-ai-automation | 1. Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | Prism documents automated campaign and ad-set budget changes through Smartly, with platform-specific update cadences. I found no callable Fospha MCP or API operation executing those ad-platform budget changes directly. | https://help-center.fospha.com/help-center/prism-fospha-smartly-how-to-set-up-a-smartly-pba-test-using-fospha-data | 1. Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.25 | The AI and Automation page describes ad-platform actions and bidding integration at product level. I found no specific MCP or API bid-parameter mutation, and the technical Prism guide documents budget allocation rather than bidding changes. | https://www.fospha.com/platform/fospha-ai-automation https://help-center.fospha.com/help-center/prism-fospha-smartly-how-to-set-up-a-smartly-pba-test-using-fospha-data | 1. Product page 2. Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.75 | MCP combines campaign drill-down with period comparisons of revenue and spend. I found no automatic linkage to each executed bid change. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.5 | Fospha’s managed geo-lift service delivers test results and recommendations, with a published customer validation example. I found no MCP or API operation retrieving completed geographic test results. | https://www.fospha.com/platform/incrementality https://www.fospha.com/case-studies/galeries-lafayette | 1. Product page 2. Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0 | Published incrementality materials describe geographic advertising experiments. I found no owned-media randomized or controlled test-result reporting through MCP, API or the broader platform. | https://www.fospha.com/platform/incrementality | 1. Product page |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.25 | Fospha publishes a CarParts example validated through a Meta conversion lift study. I found no documented platform reporting workflow or MCP/API retrieval for those studies, so the evidence is limited to the broader offering. | https://www.fospha.com/fospha-academy-lessons/incrementality-testing-what-it-is-and-how-it-compares-to-always-on-measurement | 1. Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | Fospha’s measurement scientists prioritize business questions and align test timing with media strategy. I found no MCP or API experiment-prioritization tool, or documented uncertainty-based selection process in the platform. | https://www.fospha.com/platform/incrementality | 1. Product page |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.25 | Fospha documents a managed test-design engine that selects matched treatment and control regions. I found no MCP or API design tool, or complete platform output specifying power, duration and required spend. | https://www.fospha.com/platform/incrementality | 1. Product page |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | The technical guide documents Claude and ChatGPT connection and authentication. This takes precedence over older product FAQs describing ChatGPT as forthcoming. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | Published MCP skills ask the client to render Markdown. I found no documented server-provided chart or table artifact for native display. | https://help-center.fospha.com/help-center/ask-fospha-ai-skill-library | 1. Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | The published library names MCP tools and distinguishes attribution from incremental metrics. API references also define metrics, dimensions and typed results. | https://help-center.fospha.com/help-center/ask-fospha-ai-skill-library https://api-docs.uk.fospha.com/api/marketing-performance/types/enums/marketing-metric | 1. Technical doc 2. Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The API returns structured records, aggregates, pagination and query metadata. Documented authenticated retrieval supports reuse in downstream reporting systems. | https://api-docs.uk.fospha.com/guides/retrieve-data https://api-docs.uk.fospha.com/api/marketing-performance/operations/queries/marketing-aggregate | 1. Technical doc 2. Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 1 | Fospha publishes repeatable MCP workflows with named calls and Claude Code installation instructions. These support recurring external-agent analysis, with scheduling supplied by the host runtime. | https://help-center.fospha.com/help-center/ask-fospha-ai-skill-library | 1. Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | Beam uses Bayesian inference for response curves, while Core’s technical description includes an ensemble and XGBoost. Model-backed answers are supported, but a Bayesian MMM backbone for all MCP recommendations is only partially evidenced. | https://help-center.fospha.com/help-center/beam-incremental-forecasting https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide | 1. Technical doc 2. Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 0.5 | The incrementality product describes feeding experiments into MMM and forecasts, while Enterprise pricing labels lift calibration beta. The release page still labels Lift Test Calibration as Next, so current calibration support is only partially evidenced. | https://www.fospha.com/platform/incrementality https://www.fospha.com/pricing https://www.fospha.com/whats-new | 1. Product page 2. Product page 3. Product page |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | Model Spotlight documents validation metrics including NRMSE, RMSE and testing on unseen data. Core explicitly states that validation metrics are visible in the platform, meeting the platform-level scoring rule. | https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide https://www.fospha.com/platform/core | 1. Technical doc 2. Product page |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | Users can inspect model stages and edit sale-period exclusions that affect forecasting. I found no self-service editing and audit history for Bayesian priors or the full model configuration. | https://help-center.fospha.com/help-center/understanding-your-fospha-model-complete-guide https://help-center.fospha.com/help-center/how-to-use-annotations-in-fospha | 1. Technical doc 2. Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | Published API operations provide data access and forecasting summaries. I found no controlled model-calibration mutation with validation and version history through MCP or API. | https://api-docs.uk.fospha.com/guides/retrieve-a-token https://api-docs.uk.fospha.com/api/incremental-forecasting/operations/queries/incremental-forecasting-summary | 1. Technical doc 2. Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Fospha’s public trust center names 15 customers, including Lululemon, Dyson and Urban Outfitters. I could not substantiate ten individually qualifying $1B-plus revenue brands from permitted sources, so the enterprise-reference threshold is only partially evidenced. | https://trust.fospha.com/trusted-by | 1. Technical doc |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Fospha’s public trust center marks ISO 27001 v2022 as certified. SOC 2 is labeled coming soon, so this score rests on ISO certification rather than SOC 2. | https://trust.fospha.com/ | 1. Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0.5 | Technical documentation offers UK and US regional endpoints, and the privacy policy identifies London hosting. I found no explicit customer choice of US versus EU storage, and UK residency does not establish EU residency. | https://help-center.fospha.com/help-center/fospha-api-programmatic-access-to-your-performance-data https://www.fospha.com/application-privacy-policy | 1. Technical doc 2. Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | The privacy policy identifies AWS hosting and other cloud subprocessors. I found no customer-selectable deployment option spanning AWS, GCP and Azure. | https://www.fospha.com/application-privacy-policy | 1. Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | The MCP guide explicitly supports Fospha SSO authentication. This is direct evidence of enterprise single sign-on support. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | The guide requires an onboarded Fospha account. I found no ungated, publicly promptable MCP trial or demo. | https://help-center.fospha.com/help-center/fospha-mcp-setup-onboarding-guide | 1. Technical doc |
Haus
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.75 | Haus says its MCP makes experiment, MMM, Causal Attribution, KPI and spend data available to a brand's LLM tools, and the platform measures DTC revenue and orders via Shopify and Amazon integrations. This is a marketing statement only; we found no public MCP documentation confirming online sales reporting for a specified period. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.5 | We found no evidence that the MCP reports physical-store sales specifically. The platform ingests retail and marketplace sales, offers data warehouse integrations (e.g. Snowflake, BigQuery) and a Retail breakouts add-on, so offline sales data exists in the broader platform. | https://www.haus.io/pricing | Product page |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 0.75 | The MCP is described as exposing spend data, and Haus integrates with major ad platforms such as Meta, Google and TikTok for experiments and Causal Attribution. We found no MCP documentation listing which media metrics beyond spend are returned. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.5 | We found no evidence that the MCP reports offline media spend or metrics specifically. Causal MMM tracks linear TV, out-of-home, influencer and podcast, and Fixed Geo Tests cover OOH, regional radio and direct mail, so offline media data is used in the broader platform; print is not named. | https://www.haus.io/causal-mmm | Product page |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 0.75 | The MCP is described as exposing KPI data, and the platform measures custom business metrics such as LTV, new vs. returning customers and orders. We found no MCP documentation listing the supported outcome metrics. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.25 | We found no evidence of MCP filters for date, brand, market, product and campaign. The platform offers ad-level Causal Attribution, multi-brand and international add-ons, but we found no evidence of product-level filtering. | https://www.haus.io/pricing | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 0.75 | The MCP is described as exposing Causal Attribution, experiment and MMM data, and Causal Attribution produces incrementality-calibrated ROAS and revenue per channel, campaign and ad. The MCP claim comes from marketing articles, and we found no MCP documentation of the returned fields. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.75 | The MCP is described as exposing MMM and experiment data, and Causal MMM measures hard-to-test channels such as linear TV, OOH, influencer and podcast. We found no MCP documentation confirming offline iROAS outputs specifically. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.25 | We found no evidence that the MCP or platform reports promotion-driven revenue as a separate contribution. Causal MMM offers time-varying efficiency analytics for sale periods and Architect takes promotions into account as context, which only partly covers this. | https://www.haus.io/causal-mmm | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.25 | Haus states that Causal MMM refreshes weekly, and we found no evidence of daily MMM-based iROAS. Daily incrementality reporting exists through Causal Attribution, which calibrates attribution with experiments rather than the MMM. | https://www.haus.io/causal-mmm | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.25 | We found no evidence that the MCP exposes a base, media and non-media decomposition. Causal MMM evaluates past channel mix performance and accounts for seasonality, but we found no explicit description of baseline or non-media driver contributions. | https://www.haus.io/causal-mmm | Product page |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | We found no evidence that the MCP invokes an optimization engine. Causal MMM lets users specify a time period and budget and recommends where to dial spend up or down, projecting channel-level efficiency and total KPI outcomes. | https://www.haus.io/blog/causal-intelligence-how-ai-works-in-haus | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.5 | We found no evidence of MCP access to forecasting. Causal MMM's Budget Playground and scenario planning simulate outcomes for budget allocations in the platform. | https://www.haus.io/blog/getting-started-with-causal-mmm | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | The MCP is described as exposing MMM data, and Causal MMM provides saturation analysis and return curves anchored by experiments. We found no MCP documentation confirming that response curves or marginal returns are returned. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.5 | We found no evidence of budget simulation through the MCP. In the platform, Causal MMM answers what-if questions such as moving $1M from Meta to YouTube, and Architect runs what-if analyses. | https://www.haus.io/blog/getting-started-with-causal-mmm | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.25 | We found no evidence that the MCP passes planning constraints. The platform describes planning by time period and budget and guardrails such as spending limits in Architect, but we found no documented channel-level constraints for optimization. | https://www.haus.io/blog/causal-intelligence-how-ai-works-in-haus | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.25 | We found no evidence of MCP scenario comparison against a baseline. Causal MMM supports scenario planning to simulate budget shifts, but we found no explicit description of comparing scenarios with a baseline plan. | https://www.haus.io/pricing | Product page |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.25 | We found no evidence of MCP planning against a business target. Causal MMM projects total KPI outcomes for a given budget and Architect considers CAC targets as context, which only partly covers target-based planning. | https://www.haus.io/blog/causal-intelligence-how-ai-works-in-haus | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0 | We found no evidence of saved plan retrieval through the MCP or of saved scenarios in the broader platform. The public materials do not describe scenario storage with identifiers, inputs and outputs. | https://www.haus.io/causal-mmm | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The MCP is described as exposing Causal Attribution data, which provides incrementality-calibrated ROAS, CPA, orders and revenue down to the ad level. The MCP claim is in marketing articles only, and we found no documentation of campaign or ad set fields. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.5 | We found no evidence that the MCP returns last-click or platform ROAS side by side with incremental ROAS. Causal Attribution in the platform offers first-click, last-click and linear models for comparison with the incrementality-calibrated model. | https://www.haus.io/blog/introducing-causal-attribution-your-new-daily-incrementality-solution | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.25 | We found no evidence of campaign or ad set marginal returns through the MCP. Architect shows where campaigns and ad sets are oversaturated or undersaturated, which only partly covers modeled marginal returns. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | We found no evidence of budget recommendations through the MCP. Architect recommends the next best action broken down to the specific campaigns and ad sets to move money from and to, and the homepage shows a daily shift example. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | We found no evidence of bid target recommendations through the MCP. Haus mentions bid strategy changes grounded in incrementality data as delegable actions, but does not explicitly describe target ROAS recommendations. | https://www.haus.io/article/what-is-agentic-marketing-measurement | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | We found no evidence that the MCP pushes budget changes. Architect pushes accepted changes to major ad platforms through API connections originally built to run experiments. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.25 | We found no evidence that the MCP pushes bidding changes. Architect pushes accepted changes to ad platforms and Haus lists bid strategy changes among delegable actions, but bid parameter execution is not described explicitly. | https://www.haus.io/article/what-is-agentic-marketing-measurement | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | We found no evidence of pre/post analysis of executed changes through the MCP. Haus describes a reporting agent that tracks whether changes worked and feeds results back, without a documented campaign-level pre/post comparison. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | The MCP is described as making experiment data available to LLM tools, and Haus runs GeoLift and Fixed Geo Tests. This comes from marketing articles, and we found no MCP documentation of the returned test fields. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.75 | The MCP is described as exposing experiment data, and Haus runs retail A/B tests (sampling, merchandising) and Fixed Geo Tests for direct mail. We found no MCP documentation and no evidence covering email tests specifically. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.25 | We found no evidence that the MCP or platform retrieves Meta Conversion Lift results. Haus says that channels without a Haus test can use trusted third-party studies as informed MMM starting points, which only partly covers this. | https://www.haus.io/blog/why-incrementality-testing-belongs-with-your-mmm-and-where-to-start | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.5 | We found no evidence of experiment prioritization through the MCP. Causal MMM recommends new experiments in order of importance based on goals and budget, and says when to rerun a test. | https://www.haus.io/blog/getting-started-with-causal-mmm | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | We found no evidence of test design through the MCP. The platform configures geo experiments with on-demand power analyses, synthetic controls and Copilot-suggested test setups. | https://www.haus.io/science | Product page |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 0.5 | Haus states that its MCP connects experiment, MMM, attribution, KPI and spend data to a brand's preferred LLM tools. We found no public connection or authentication instructions for Claude or ChatGPT; the support knowledge base content is not publicly visible. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | We found no evidence of server-provided tables or chart artifacts from the MCP. The public materials do not describe MCP output formats. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0.5 | The Haus Knowledge Base has a Haus API collection, which suggests API documentation exists. Its articles are not publicly visible, and we found no public MCP tool definitions or metric documentation. | https://support.haus.io/collections/2290813438-haus_api | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 0.5 | The MCP is described as returning experiment, MMM, attribution, KPI and spend data, and a Haus API collection exists in the knowledge base. We found no public documentation of structured result formats or downloadable files. | https://support.haus.io/collections/2290813438-haus_api | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0 | We found no evidence of documented MCP invocation from an external agent runtime for recurring workflows. Haus's own Architect agent runs inside the platform. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | Haus describes its Causal MMM approach in terms of Bayesian priors, with experiments treated as ground truth and incrementality priors for untested channels, and the MCP exposes MMM data. The site does not state explicitly that Causal MMM is a Bayesian model or name the estimation method. | https://www.haus.io/blog/why-incrementality-testing-belongs-with-your-mmm-and-where-to-start | Marketing collateral |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Causal MMM uses incrementality experiments as ground truth that anchor the response curve, and every new test automatically feeds back into the model. For untested channels, Haus priors tailored to the brand's own experiments are used. | https://www.haus.io/blog/why-incrementality-testing-belongs-with-your-mmm-and-where-to-start | Marketing collateral |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 0.5 | We found no evidence that Haus reports MMM validation metrics such as R2, MAPE or holdout error. For experiments, Haus runs thousands of placebo tests and shows confidence intervals, which only partly covers this. | https://www.haus.io/blog/causal-intelligence-how-ai-works-in-haus | Marketing collateral |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | Haus says Causal MMM visibly shows the tests calibrating the model so users know how data drives recommendations. We found no evidence that users can edit priors or model settings in a self-serve UI; plans include an MMM Specialist. | https://www.haus.io/blog/a-first-look-at-causal-mmm | Marketing collateral |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We found no evidence that agents can propose or apply calibration changes through the MCP. The public materials describe the MCP as a way to read data only. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 1 | The site shows logos and stories for brands including Intuit, Wayfair, Dyson, SharkNinja, Invisalign, FanDuel, Sonos, Grubhub, Jameson and Oura, many with $1B+ revenue. Haus also states that its customers have a combined revenue of $1T+. | https://www.haus.io/customers | Product page |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | The Haus Trust Center lists SOC 2 Type 2 and ISO/IEC 27001:2022 compliance, with the SOC 2 report available on request. The website footer shows SOC 2 Type 2 (Sensiba) and ISO badges. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | We found no evidence of a choice between US and EU data residency. The Trust Center lists Google Cloud as the infrastructure subprocessor without a region option. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We found no evidence of a choice between AWS, GCP and Azure. The Trust Center names Google Cloud as the cloud infrastructure subprocessor. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | We found no evidence of enterprise SSO (SAML, Okta, Microsoft Entra). The Trust Center lists multi-factor authentication but not single sign-on. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We found no evidence of a public, promptable MCP demo or trial. All plans on the pricing page route to 'Get a demo', and the only public interactive asset found is a case study. | https://www.haus.io/pricing | Product page |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.5 | Time Tests compare actual revenue during a chosen campaign window with a forecast, and Haus supports DTC sales measurement. I found no specific MCP or API documentation for retrieving actual online sales by period. | https://www.haus.io/blog/measuring-big-brand-moments-with-time-tests https://www.haus.io/use-cases/amazon-and-retail | Marketing collateral Product page |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.25 | Haus ingests retail sales data and measures lift across physical retail and other sales channels. I found no documentation of warehouse-sourced store-sales reporting for a requested period, or its availability through MCP or API. | https://www.haus.io/use-cases/amazon-and-retail | Product page |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 0.75 | Haus describes MCP access to spend data alongside its cross-channel Causal Attribution reporting. This is partial MCP evidence because I found no tool schema specifying supported platforms and media metrics. | https://www.haus.io/article/what-is-agentic-marketing-attribution https://www.haus.io/causal-attribution | Marketing collateral Product page |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.25 | Haus supports experiments for linear TV, out-of-home, regional radio and direct mail. This only partially supports offline-media reporting, as I found no spend-and-media-metric report covering TV, OOH, radio and print through the platform, MCP or API. | https://www.haus.io/experiments | Product page |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 0.75 | Haus describes MCP access to KPI data, and Causal Attribution explicitly reports orders and acquisition metrics beyond revenue. I found no MCP or API schema confirming the exact supported outcome fields, so the MCP evidence is partial. | https://www.haus.io/article/what-is-agentic-marketing-attribution https://www.haus.io/blog/introducing-causal-attribution-your-new-daily-incrementality-solution | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.25 | Haus documents configurable channel groupings and campaign-level analysis in its broader platform. I found no complete date, brand, market, product and campaign filtering contract or corresponding MCP or API parameters. | https://www.haus.io/blog/why-your-mmm-disagrees-with-how-you-actually-plan https://www.haus.io/causal-attribution | Marketing collateral Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 0.75 | Haus describes MCP access to Causal Attribution data, which includes experiment-calibrated ROAS and revenue. I found no public result schema confirming both incremental measures for each digital channel, so this is partial MCP evidence. | https://www.haus.io/article/what-is-agentic-marketing-attribution https://www.haus.io/blog/introducing-causal-attribution-your-new-daily-incrementality-solution | Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.25 | Causal MMM measures hard-to-test offline channels including linear TV and out-of-home, while Haus experiments also cover regional radio. I found no explicit channel-by-channel offline iROAS and incremental-revenue output, or MCP or API access to those fields. | https://www.haus.io/causal-mmm https://www.haus.io/experiments | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.25 | Haus supports analysis during promotional periods and says its modelling extends measurement to external business factors. I found no explicit promotion-and-pricing revenue decomposition or corresponding MCP or API output, leaving only partial platform evidence. | https://www.haus.io/use-cases/holidays-and-promotions https://www.haus.io/blog/geolift-mmm-getting-more-from-your-incrementality-practice | Product page Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0 | Haus explicitly advertises weekly Causal MMM refreshes, while daily refreshes belong to Causal Attribution. I found no evidence of daily MMM-based incremental ROAS measurement through MCP, API or the broader platform. | https://www.haus.io/causal-mmm https://www.haus.io/causal-attribution | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.25 | Haus positions Causal MMM as connecting media effects with external factors such as seasonality and pricing. I found no explicit output decomposing baseline, media and non-media contributions, or MCP or API access to such a decomposition. | https://www.haus.io/blog/geolift-mmm-getting-more-from-your-incrementality-practice | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | Haus describes Causal MMM as an optimization engine that guides budget allocation across channels. I found no documented MCP or API operation invoking that engine with an objective and planning period. | https://www.haus.io/blog/geolift-mmm-getting-more-from-your-incrementality-practice | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.5 | Causal MMM lets users move spend across channels and simulate resulting outcomes. I found no documented MCP or API forecast operation accepting a specified media plan. | https://www.haus.io/causal-mmm | Product page |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.5 | Haus documents Causal MMM response curves and marginal-return analysis for channel diversification. I found no MCP or API operation returning those curves or marginal-return values. | https://www.haus.io/blog/geolift-mmm-getting-more-from-your-incrementality-practice | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.5 | Causal MMM explicitly supports what-if analysis of budget shifts across channels. I found no documented MCP or API operation calculating a user-specified spend change. | https://www.haus.io/causal-mmm | Product page |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.25 | Architect is described as adapting recommendations to business constraints and promotions. I found no complete implementation evidence for user-defined budget limits, channel restrictions and planning dates, or their enforcement through MCP or API. | https://www.haus.io/blog/can-an-ai-agent-make-budget-decisions-youd-bet-your-business-on | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.25 | Causal MMM supports budget scenarios and their predicted outcomes. I found no explicit comparison output containing both alternative scenarios and a named baseline plan, or corresponding MCP or API access. | https://www.haus.io/causal-mmm | Product page |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.25 | The StockX case shows Causal MMM recommendations evaluated against a profitability iROAS target. I found no documented target-solving budget planner accepting an outcome level, or an MCP or API operation exposing one. | https://www.haus.io/customer/stockx | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0 | The reviewed Causal MMM material describes scenario simulation but does not document retrieval of saved plans. I found no evidence of saved scenario identifiers, assumptions and outputs being retrievable through the platform, MCP or API. | https://www.haus.io/causal-mmm | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | Haus describes MCP access to Causal Attribution, whose launch material specifies calibrated ROAS and revenue down to individual ads. I found no MCP schema confirming campaign and ad-set result granularity, so this remains partial MCP evidence. | https://www.haus.io/article/what-is-agentic-marketing-attribution https://www.haus.io/blog/introducing-causal-attribution-your-new-daily-incrementality-solution | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.5 | Causal Attribution offers incrementality-calibrated metrics down to ads, comparisons with last-click models and adjusted platform reporting. I found no MCP or API documentation for retrieving the requested campaign and ad-set comparisons. | https://www.haus.io/blog/introducing-causal-attribution-your-new-daily-incrementality-solution https://www.haus.io/causal-attribution | Marketing collateral Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.25 | Architect identifies campaign and ad-set opportunities using over- and undersaturation signals. I found no explicit numerical marginal-return output for each campaign and ad set, or an MCP or API operation returning it. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | Architect recommends budget movements among specific campaigns and ad sets. I found no evidence that it returns optimal daily spend amounts for each entity through the broader platform, MCP or API. | https://www.haus.io/article/what-is-agentic-marketing-attribution | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | Haus explains how incrementality factors translate business acquisition goals into platform CPA targets. I found no product evidence of optimal bid recommendations for each campaign and ad set, or corresponding MCP or API output. | https://www.haus.io/blog/incrementality-fundamentals | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.75 | Haus describes execution agents pushing approved budget changes to major advertising platforms through API connections. The evidence is partial because I found no endpoint documentation specifying daily budget fields, entity coverage or supported platforms. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | The reviewed agentic-media material describes API execution of approved changes but does not specify bid parameters. I found no evidence of Haus pushing target ROAS or other bidding changes through MCP, API or its broader platform. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | Haus describes reporting agents tracking outcomes after approved changes and feeding results into subsequent recommendations. I found no explicit campaign or ad-set pre/post revenue-and-spend report tied to each executed bid change, including through MCP or API. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | Haus documents geographic experiment readouts and describes MCP access to experiment data. I found no public retrieval schema confirming completed tests with their intervention and outcome fields, so MCP support is only partially evidenced. | https://www.haus.io/experiments https://www.haus.io/article/what-is-agentic-marketing-attribution | Product page Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.5 | Haus Fixed Geo Tests cover direct mail, and its retail offering includes controlled sampling and merchandising tests. I found no MCP or API operation retrieving these owned-media incrementality results. | https://www.haus.io/experiments https://www.haus.io/use-cases/amazon-and-retail | Product page |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.25 | The Sunday case confirms that Haus incorporates Meta Conversion Lift studies into Causal MMM response curves. I found no evidence of retrieving the original study findings through a dedicated platform view, MCP or API, so support is partial. | https://www.haus.io/customer/sunday | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.5 | Causal MMM identifies new testing opportunities, and the StockX case shows a specific Criteo scaling hypothesis being tested. I found no MCP or API operation prioritizing experiments from measurement results and uncertainty. | https://www.haus.io/causal-mmm https://www.haus.io/customer/stockx | Product page Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | Haus documents test/control assignment and an in-platform calculator for power, holdout size and duration, with spend informing feasibility. I found no MCP or API operation producing this experiment design. | https://www.haus.io/experiments https://www.haus.io/blog/incrementality-fundamentals | Product page Marketing collateral |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 0 | Haus publicly mentions connecting causal data to preferred LLM tools through MCP. I found no public connection and authentication instructions covering both Claude and ChatGPT. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | Haus product pages show platform charts and tables, but these do not establish native MCP visual artifacts. I found no documented compatible client displaying server-provided chart or table artifacts. | https://www.haus.io/causal-attribution | Product page |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0.5 | Haus publishes a glossary explaining measurement metrics and methods. I found no discoverable MCP tool definitions or field-level response documentation, so the combined requirement is only partially evidenced. | https://www.haus.io/glossary | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 0.5 | Haus describes making measurement data available to external LLM tools and its terms include downloadable or exportable outputs. I found no public structured-result schema or complete downloadable dataset example, so evidence of reusable structured data is partial. | https://www.haus.io/article/what-is-agentic-media-buying https://www.haus.io/legal/terms | Marketing collateral Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0 | Haus describes agentic workflows and MCP access at a conceptual level. I found no documented repeatable MCP invocation from an external agent runtime for scheduled or recurring analysis. | https://www.haus.io/article/what-is-agentic-media-buying | Marketing collateral |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | Haus grounds Causal MMM recommendations in experiments and describes proprietary algorithms that go beyond standard Bayesian methods. I found no explicit confirmation of the required Bayesian backbone and deterministic answer contract, so the combined requirement has partial evidence. | https://www.haus.io/blog/mmm-software-what-should-you-look-for | Marketing collateral |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Causal MMM uses experiments as causal anchors and automatically incorporates new completed tests into the model. The Sunday case additionally documents calibration using both Haus GeoLift and Meta Conversion Lift studies. | https://www.haus.io/causal-mmm https://www.haus.io/customer/sunday | Product page Marketing collateral |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 0.5 | Haus documents its Model Reliability Index, placebo validation and automated model-quality selection. I found no explicit customer-facing MMM validation report exposing these diagnostics or the requested modelling KPIs, so support is partial. | https://www.haus.io/blog/how-haus-scales-causal-marketing-measurement-without-human-bias | Marketing collateral |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0 | Haus discusses prior assumptions and model transparency in its methodology material. I found no evidence of a self-serve interface where customers can both audit and edit priors or other key model settings. | https://www.haus.io/blog/trust-in-trust-out-why-an-mmm-built-on-experiments-yields-more-accurate-results | Marketing collateral |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | Haus documents automatic experiment integration and internal controls over model configuration. I found no authorized-agent calibration workflow with validation and version history through MCP or an equivalent API. | https://www.haus.io/blog/how-haus-scales-causal-marketing-measurement-without-human-bias https://www.haus.io/blog/mmm-software-what-should-you-look-for | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Haus publicly names enterprise customers including FanDuel, Intuit, Dyson, Wayfair, Sonos and SharkNinja. I could not verify at least ten named reference brands each exceeding one billion dollars in revenue using the permitted sources, so the threshold is only partially supported. | https://www.haus.io/ | Product page |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Haus Trust Center lists SOC 2 Type 2 and ISO/IEC 27001:2022 and identifies SOC 2 and penetration-testing reports. This is strong public evidence of the required independent security assurance, although the underlying reports require access approval. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | Haus documents safeguards for international transfers, including EEA-to-US transfers. I found no public evidence that customers can choose between US and EU data residency. | https://www.haus.io/legal/data-processing-addendum | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | Haus Trust Center identifies Google Cloud as a cloud infrastructure subprocessor. I found no public evidence of customer-selectable deployment across AWS, GCP and Azure. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | The public Trust Center lists authentication-related controls, but detailed product-security information requires access approval. I found no public confirmation of customer-facing enterprise SSO for the Haus marketing platform. | https://trust.haus.io/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | Haus directs prospects to a demo request form and sales-led product walkthroughs. I found no publicly accessible, promptable MCP demo or trial available without a sales call. | https://www.haus.io/demo https://www.haus.io/blog/introducing-causal-attribution-your-new-daily-incrementality-solution | Product page Marketing collateral |
Lifesight
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.75 | Lifesight's MCP documentation says the connector gives Claude and ChatGPT real-time access to Marketing Mix Models, budget data and ad performance, and documented prompts ask for last month's or last quarter's revenue performance. We did not find a documented tool that returns actual online sales totals for a chosen period, so the evidence is partial. | https://docs.lifesight.io/docs/mcp-connect | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.5 | We found no evidence that the MCP reports physical-store sales. The broader platform ingests customer-selected tables from BigQuery and Snowflake data warehouses for MMM and reporting, and Lifesight states it supports offline conversions such as in-store purchases. | https://docs.lifesight.io/docs/bigquery | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The MCP docs state it provides access to budget data and ad performance, and documented prompts compare channel performance over the last N days and return platform-reported ROAS by ad set. Native integrations cover Meta, Google Ads, TikTok, Snapchat, Pinterest, Amazon Ads and other major paid platforms. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.5 | We found no evidence that the MCP returns spend or media metrics for TV, out-of-home, radio or print. The platform's MMM accepts offline inputs such as Linear_TV_Spend and CTV_Spend, and Lifesight states it can measure CTV, OOH and other offline channels. | https://docs.lifesight.io/docs/mmm-input-schema | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | Documented MCP prompts return incremental CPA, CAC, gross margin impact, margin contribution, payback and the lifetime value of acquired cohorts alongside revenue. The P&L Translator skill on the MCP product page frames results as margin contribution, working capital impact and payback. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | MCP prompts filter by date window (last week, last N days, a quarter), channel, campaign and ad set, and the MCP reads every model in the workspace, including custom models. Brand, market and product views exist in the platform through separate geography and product models, but we found no explicit MCP documentation of those filters. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The MCP setup guide lists "What was my best-performing channel last week by iROAS?" as a verification prompt that returns a ranked answer from the customer's causal model. Other documented prompts return iROAS and incremental revenue by channel compared with platform-reported ROAS. | https://docs.lifesight.io/docs/setup-guide | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.75 | The MCP returns channel-level iROAS and incremental revenue from the workspace's MMM, and the platform's MMM accepts offline inputs such as linear TV, CTV and OOH. We did not find an MCP example that names an offline channel, so the evidence is partial. | https://docs.lifesight.io/docs/mcp-connect | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.75 | Documented MCP prompts simulate a sitewide discount against brand investment, compare three BFCM discount depths and ask whether a channel spike was correlated with a promotion. The platform's Event Impact Summary isolates the impact of past promotions, but we found no MCP example that reports historical promotion-driven revenue as its own line. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.25 | We found no evidence of daily MMM-based iROAS in the MCP, and Lifesight's documentation describes a weekly model refresh with quarterly retraining. Causal attribution applies MMM-derived incrementality factors to near-real-time platform data, which is partial evidence of more frequent MMM-based iROAS in the broader platform. | https://docs.lifesight.io/docs/model-refresh | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | An MCP prompt returns a causal decomposition of a channel's change into incremental contribution, cannibalization from adjacent channels and external-factor contribution, and other prompts project the baseline. The underlying model reports baseline, paid, organic and contextual contributions in its Insights tab. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.75 | The MCP overview says users can ask "Optimize my Q4 budget for revenue" instead of running scenarios in the platform, and documented prompts ask where to reallocate a given budget to maximize revenue. The MCP product page FAQ lists scenario simulation skills as coming next, so it is unclear whether the MCP invokes the Planner's optimizer directly. | https://docs.lifesight.io/docs/mcp-connect | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.75 | Documented MCP prompts project incremental revenue and CAC for the current BFCM plan and project the baseline under a specified cut to upper-funnel spend. We did not find documentation that the MCP passes a user-defined channel allocation to the Planner's forecasting engine. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | An MCP prompt asks at what spend level a channel saturates and returns a spend-versus-iROAS curve description, a saturation point and a recommended cap. The evidence is partial because we found no documented MCP output of mROAS values or response-curve data points, which the platform shows in its Insights tab. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.75 | Documented prompts ask what moving a given percentage of spend to higher-iROAS channels would do to incremental revenue and CAC, and the Scenario Planner skill compares a plan with +15% and -15% budgets. The same product page FAQ lists scenario simulation skills as a coming feature, so the evidence is partial. | https://lifesight.io/mcp/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.5 | We found no evidence that the MCP passes budget limits or channel restrictions to the optimizer. The platform Planner supports preset and manual lower and upper spend limits per channel, forecast periods and custom pacing. | https://docs.lifesight.io/docs/build-a-media-plan-copy | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.75 | The MCP Scenario Planner skill compares a current plan against +15% and -15% budgets and projects the outcome of each, and a documented prompt returns status quo, increase and decrease scenarios. The same page lists scenario simulation skills as coming next, so it is unclear whether these comparisons run on the Planner's scenarios. | https://lifesight.io/mcp/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.75 | MCP prompts ask for the budget range that maximizes incremental revenue next year and the brand investment needed to raise the baseline revenue floor by a given percentage. The platform Planner has a Target KPI goal for revenue or orders, but we did not find documentation that the MCP calls it. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0.75 | The MCP prompt library refers to a save_insight tool and saved bookmarks that can be re-pulled later, and a suggested prompt returns the budget allocation by channel for a quarter. We did not find documentation that the MCP retrieves Planner scenarios with their identifiers, inputs and outputs. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 1 | Documented MCP prompts return Lifesight iROAS by ad set and incremental CPA for a named campaign, and the MCP launch post lists "Which campaigns are delivering the highest incremental return?" as a typical question. The platform's causal attribution computes incremental revenue at campaign, ad set and ad level. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | An MCP prompt returns platform-reported ROAS against Lifesight iROAS by ad set and flags gaps greater than 2x. We found no MCP example that adds last-click ROAS, although the platform supports last-touch attribution models. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.5 | We found no evidence that the MCP returns miROAS for campaigns or ad sets. The platform's causal attribution computes miROAS at tactic level and scales it down to campaign and ad set level. | https://docs.lifesight.io/docs/causal-attribution | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | We found no evidence that the MCP returns daily budget recommendations. The platform Optimizer translates the default scenario into daily or weekly budget changes for individual campaigns and ad sets. | https://docs.lifesight.io/docs/automated-budget-optimization | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | We found no evidence that the MCP recommends bid values. The platform Optimizer shows editable Bid Strategy values next to recommended budgets, but we found no documentation of recommended target ROAS values. | https://docs.lifesight.io/docs/manual-changes | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | We found no evidence that the MCP can push budget changes, and the prompt library (updated 2026-06-03) states all prompts are read-only with write actions planned for Q3. The platform Optimizer applies selected budget changes directly to campaigns on live ad platforms. | https://docs.lifesight.io/docs/automated-budget-optimization | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.5 | We found no evidence that the MCP can push bidding changes. The platform Optimizer lets users edit bid amounts for campaigns and ad sets across ad platforms, and its Logger records Bid Strategy changes. | https://docs.lifesight.io/docs/manual-changes | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | We found no evidence of MCP pre/post analysis for executed changes. The platform Logger records every budget and bid change with prior and new values so users can track performance shifts, but we found no documented pre/post impact summary. | https://docs.lifesight.io/docs/change-logs | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | An MCP prompt retrieves the holdout setup of the most recent geo test, including markets, duration and confidence level, and the search_knowledge_base tool covers past experiments. We did not find an MCP example that returns the measured lift, although the platform's geo experiment results include adjusted lift and significance. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.25 | We found no evidence that the MCP reports owned-media test results. The platform supports split tests and time tests and lists A/B testing as a calibration input, but we found no documentation of email or leaflet test results. | https://docs.lifesight.io/docs/calibration | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.25 | We found no evidence that the MCP reports Meta Conversion Lift findings. The platform's calibration methodology accepts platform conversion lift studies from Facebook, Google and YouTube as inputs, which is partial evidence that those results are held in the platform. | https://docs.lifesight.io/docs/calibration | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.5 | We found no evidence that the MCP recommends which experiments to run. The platform's Attribute Quality Score flags channels with wide ROI confidence intervals for testing, and Recommendations include experimentation recommendations with expected lift and cost. | https://docs.lifesight.io/docs/calibration | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | We found no evidence that the MCP designs incrementality tests. The platform's geo experiment designer proposes test and synthetic control markets with duration, additional spend and minimum detectable lift, and shows a power curve. | https://docs.lifesight.io/docs/geo-experiment-creation-1 | Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Lifesight publishes step-by-step setup guides for Claude Desktop (custom connector with OAuth) and ChatGPT Codex (Streamable HTTP with a personal access token). Both guides give the server URL and verification prompts. | https://docs.lifesight.io/docs/setup-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0.5 | The setup guide says the model-inventory prompt returns a Markdown table, which is partial evidence of structured tabular output. We found no evidence of server-provided chart artifacts, and the prompt library describes a "projection chart description" rather than a chart. | https://docs.lifesight.io/docs/setup-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | The prompt library names MCP tools (list_models, search_knowledge_base, save_insight), and the MCP includes documentation search over Lifesight's methodology library. Answers include methodology footnotes and audit-trail details such as model name, run timestamp and confidence interval. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 0.5 | The MCP returns structured, model-backed answers such as ranked tables that the docs suggest pasting into planning documents. We found no evidence of complete structured datasets or downloadable files through the MCP. | https://docs.lifesight.io/docs/mcp-connect | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | The ChatGPT Codex guide documents bearer-token access to the MCP server, which allows use from an agent runtime, and prompts suggest saving recurring outputs such as a morning briefing. We found no documentation of scheduled or programmatic recurring invocation, and Cursor and Claude Code support is listed as roadmap. | https://docs.lifesight.io/docs/setup-guide-chatgpt | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | MCP answers are model-backed and come from Lifesight's causal MMM rather than last-click attribution. Lifesight states its inference is primarily frequentist (ridge regression with evolutionary tuning and bootstrapping), so the backbone is not a Bayesian MMM. | https://docs.lifesight.io/docs/faq | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Lifesight documents calibration of the MMM against geo-lift, split, A/B and platform conversion lift results, using holistic and contextual calibration methods. Users add calibration rows with channel, dates, ROAS and confidence in the platform UI. | https://docs.lifesight.io/docs/calibration | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | Lifesight reports training and testing R-squared, adjusted R-squared, MAPE, NRMSE, holdout accuracy, backtesting and residual diagnostics for its models. The Insights tab shows NRMSE and estimation error for the model version in use. | https://docs.lifesight.io/docs/goodness | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 1 | Lifesight states users can configure custom causal DAGs and weakly informative priors for adstock and saturation when training models in the platform. The calibration UI lets users enter ROAS, confidence and calibration type per channel and review the before-and-after impact. | https://docs.lifesight.io/docs/lifesights-approach | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We found no evidence that agents can propose or apply calibration changes through the MCP. The prompt library states the MCP is read-only, and calibration is documented only in the platform UI. | https://docs.lifesight.io/docs/prompt-library | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0 | We found no evidence of 10 public reference customers with $1B+ revenue. Named customers on the case-studies page include Seidensticker and Obvi, while most case studies are anonymized. | https://lifesight.io/case-studies/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Lifesight's documentation states the company is SOC 2 compliant and lists ISO 27001 among its certifications. The FAQs repeat that Lifesight has ISO 27001 and SOC 2 certification. | https://docs.lifesight.io/docs/soc-2 | Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 1 | Lifesight offers data localization at workspace level, with dedicated BigQuery datasets in the United States, Europe, Asia Pacific or Middle East and Africa. The region is chosen at setup and cannot be migrated later. | https://docs.lifesight.io/docs/data-localization-security-privacy-compliance | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We found no evidence of a choice between AWS, GCP and Azure. Lifesight states the platform operates on Google Cloud infrastructure. | https://docs.lifesight.io/docs/data-localization-security-privacy-compliance | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | We found no evidence of enterprise SSO (SAML, Okta or Microsoft Entra) in Lifesight's documentation. The Claude connector authenticates through OAuth with Lifesight login credentials. | https://docs.lifesight.io/docs/setup-guide | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We found no evidence of a public, promptable MCP demo. The MCP product page says the connector returns an authentication error without an active workspace and directs evaluators to book a demo for a sandbox. | https://lifesight.io/mcp/ | Product page |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.5 | The Purchases documentation shows dated order values and product revenue, supporting actual online sales reporting in the platform. I did not find an MCP tool or callable reporting API explicitly returning actual online sales for a requested period. | https://docs.lifesight.io/docs/purchases https://docs.lifesight.io/docs/mcp-connect | Technical doc Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.5 | The platform combines in-store POS data with measurement, and its BigQuery connector imports selected customer warehouse tables for reporting. I did not find explicit MCP/API retrieval of actual physical-store sales for a requested period. | https://docs.lifesight.io/docs/bigquery https://lifesight.io/solutions/retail-cpg/ | Technical doc Product page |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 0.75 | MCP documentation promises access to ad performance and budget data, while campaign documentation identifies spend, clicks and impressions. This is partial MCP evidence because I did not find a documented response schema confirming these fields across all major connected paid platforms. | https://docs.lifesight.io/docs/mcp-connect https://docs.lifesight.io/docs/campaigns https://docs.lifesight.io/docs/connect | Technical doc Technical doc Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.5 | Lifesight documents a customer implementation covering TV, radio, billboards and print, supported by platform reporting of model input spend. I did not find an MCP/API example explicitly returning raw offline media spend and delivery metrics across those channels. | https://lifesight.io/case-study/premier-auto-dealership-chain-increases-incremental-revenue/ https://docs.lifesight.io/docs/overview | Marketing collateral Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 0.75 | The MCP product page advertises margin contribution and payback reporting. This supports partial MCP coverage, while platform documentation explicitly supports orders and new customers but does not establish their MCP/API output fields. | https://lifesight.io/mcp/ https://docs.lifesight.io/docs/model-creation | Product page Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | MCP setup examples specify periods and channels. I did not find complete MCP/API filter definitions for brand, market, product and campaign dimensions, although the platform documents campaign, product and geography analysis. | https://docs.lifesight.io/docs/setup-guide https://docs.lifesight.io/docs/campaigns https://docs.lifesight.io/docs/purchases | Technical doc Technical doc Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | Lifesight explicitly advertises channel iROAS and incremental marketing revenue through MCP. Technical measurement documentation defines digital-channel incremental revenue and iROAS, providing strong public support for retrieving causal channel performance. | https://lifesight.io/mcp/ https://docs.lifesight.io/docs/insights | Product page Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.5 | Lifesight documents MMM estimates of incremental offline-channel contributions in an implementation covering TV, radio and billboards. I did not find explicit MCP/API retrieval of both incremental revenue and iROAS for offline channels. | https://lifesight.io/case-study/premier-auto-dealership-chain-increases-incremental-revenue/ https://docs.lifesight.io/docs/insights | Marketing collateral Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.75 | The MCP library includes promotion-related causal decomposition. This provides partial access evidence, supported by platform modeling of price and promotion effects, but I did not find a defined MCP/API field for historical promotion-driven revenue. | https://docs.lifesight.io/docs/prompt-library https://lifesight.io/solutions/retail-cpg/ https://docs.lifesight.io/docs/model-creation | Technical doc Product page Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.75 | Technical documentation allows daily automatic MMM refresh through Google Sheets, and MCP exposes current model data. I found no explicit guarantee that MCP iROAS updates daily for every workspace, and the optimization product page also describes weekly model updates. | https://docs.lifesight.io/docs/google-sheets-integration https://docs.lifesight.io/docs/mcp-connect https://lifesight.io/product/optimize/ | Technical doc Technical doc Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.75 | MCP examples describe causal driver decomposition. The platform separately documents baseline, paid, organic and contextual contributions, but I did not find an MCP/API schema confirming that complete decomposition in a single result. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/insights | Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.75 | The MCP overview explicitly offers revenue-based quarterly budget optimization. The underlying Planner calculates channel allocations, but the MCP product FAQ still places scenario simulation on its roadmap, leaving partial rather than unambiguous availability evidence. | https://docs.lifesight.io/docs/mcp-connect https://docs.lifesight.io/docs/build-a-media-plan-copy https://lifesight.io/mcp/ | Technical doc Technical doc Product page |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.75 | The MCP library offers current-plan forecasts. This is partial evidence because the product FAQ calls simulation forthcoming, although technical Planner documentation confirms model-based forecasts from specified allocations in the platform. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/intrepreting-a-budget-plan https://lifesight.io/mcp/ | Technical doc Technical doc Product page |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | The MCP library describes saturation-curve answers. Channel insights document marginal ROAS and model attributes document response curves, but explicit MCP/API retrieval of both numeric marginal returns and complete spend-response data was not found. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/insights https://docs.lifesight.io/docs/model | Technical doc Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.75 | MCP documentation offers percentage-change budget simulations. The platform supports scenario calculation, but the MCP product FAQ still lists simulation as forthcoming, so public evidence of current MCP availability is inconsistent. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/build-a-media-plan-copy https://lifesight.io/mcp/ | Technical doc Technical doc Product page |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.5 | The Planner documents total budget, per-channel lower and upper limits, forecast dates and a historical basis period. I did not find an MCP/API contract showing that all these user constraints can be passed to the optimizer and returned for inspection. | https://docs.lifesight.io/docs/build-a-media-plan-copy https://lifesight.io/product/optimize/ | Technical doc Product page |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.75 | MCP examples compare alternatives with a status-quo plan. The platform documents baseline and forecast metrics, but the conflicting MCP simulation roadmap prevents treating the current external capability as fully established. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/intrepreting-a-budget-plan https://lifesight.io/mcp/ | Technical doc Technical doc Product page |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.75 | The MCP overview supports optimizing a quarterly budget for revenue. The platform also accepts a Target KPI goal, but I did not find explicit MCP/API parameters for solving to a specified revenue, acquisition or profit target. | https://docs.lifesight.io/docs/mcp-connect https://docs.lifesight.io/docs/build-a-media-plan-copy | Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0.5 | The Planner saves configured scenarios and their analyses for subsequent viewing. I did not find MCP/API retrieval of saved plan identifiers together with assumptions and calculated outputs, so strong platform evidence earns platform-only credit. | https://docs.lifesight.io/docs/intrepreting-a-budget-plan https://docs.lifesight.io/docs/mcp-connect | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | MCP examples explicitly request ad-set iROAS. Causal-attribution documentation supports granular incremental revenue, but I did not find complete MCP/API output definitions for both revenue and iROAS at campaign and ad-set levels. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/causal-attribution | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | MCP examples compare ad-set iROAS with platform ROAS. I did not find the full three-way MCP/API comparison including last-click ROAS, although the platform supports last-touch and causal attribution views. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/overview-1 | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.5 | Causal-attribution documentation distributes tactic-level marginal returns to campaigns and ad sets using a shared saturation-rate assumption. I did not find explicit MCP/API retrieval of these granular marginal estimates, so this receives strong platform-only credit. | https://docs.lifesight.io/docs/causal-attribution | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | The Optimizer documentation converts channel plans into daily or weekly budget recommendations for individual campaigns and ad sets. I did not find a documented MCP/API operation returning those daily entity-level budget values. | https://docs.lifesight.io/docs/automated-budget-optimization | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | Optimizer documentation exposes bid-strategy fields and manual bid edits, providing partial evidence of bidding support in the platform. I did not find explicit generation of optimal bid targets for each campaign or ad set in either the platform or MCP/API documentation. | https://docs.lifesight.io/docs/manual-changes https://lifesight.io/product/optimize/ | Technical doc Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | The platform documents applying campaign and ad-set budget adjustments directly to connected ad platforms. I did not find a callable external MCP/API budget-write operation, and the MCP product page still describes write actions as forthcoming. | https://docs.lifesight.io/docs/automated-budget-optimization https://lifesight.io/mcp/ | Technical doc Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.5 | Manual Changes and Logger document editing bid amounts or strategies and recording executed changes on platforms such as Facebook and Google Ads. I did not find an external MCP/API bid-write interface, so this receives platform-only credit. | https://docs.lifesight.io/docs/manual-changes https://docs.lifesight.io/docs/change-logs | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | The Logger records each changed entity, parameter, prior value, new value and execution status, providing partial support for reviewing interventions. I did not find an automatic revenue-and-spend pre/post impact analysis for each executed bidding change in the platform or MCP/API. | https://docs.lifesight.io/docs/change-logs | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | MCP supports searching past geo-test methodology. Platform documentation confirms completed-test outcomes and intervention details, but I did not find a complete MCP/API result contract for returning the measured lift and tested outcome together. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/geo-experiment-insights | Technical doc Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.25 | Lifesight technical guidance describes audience split testing as an available platform method. I did not find explicit owned-media incrementality result reporting for email or leaflet tests, or corresponding MCP/API retrieval, so the evidence is partial and platform-only. | https://docs.lifesight.io/docs/how-do-you-measure-the-impact-of-a-new-channel-launch | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0 | I did not find evidence that Lifesight retrieves and reports Meta Conversion Lift study findings through MCP/API or its platform. The reviewed platform documentation covers Lifesight geo experiments and generic experiment-based calibration, which do not establish this specific capability. | https://docs.lifesight.io/docs/geo-experiment-insights https://docs.lifesight.io/docs/model-calibration | Technical doc Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.5 | Causal Flags identifies channels with unreliable model estimates and recommends targeted experiments, while Recommendations provides experimentation guidance. I did not find explicit MCP/API prioritization of a testing agenda from those uncertainty signals. | https://docs.lifesight.io/docs/causality-flags https://docs.lifesight.io/docs/recommendations | Technical doc Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | Geo Experiment Design documents treatment and control markets, detectable lift, test duration and required additional spend. I did not find a documented MCP/API operation generating that complete feasible test design. | https://docs.lifesight.io/docs/geo-experiment-creation-1 | Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Lifesight publishes separate connection guides for Claude and ChatGPT/Codex, including the server URL and account authentication steps. The ChatGPT guide specifically describes a Codex Streamable HTTP setup with a personal access token, while Claude uses a sign-in redirect. | https://docs.lifesight.io/docs/setup-guide https://docs.lifesight.io/docs/setup-guide-chatgpt | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0.5 | The setup guide promises Markdown tables in compatible clients. I did not find evidence distinguishing server-produced visual artifacts from client-generated presentation, so native MCP visual output receives partial credit. | https://docs.lifesight.io/docs/setup-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0.5 | The MCP library names list_models and search_knowledge_base. Metric definitions are documented elsewhere, but I did not find complete discoverable tool schemas, parameter definitions and response documentation. | https://docs.lifesight.io/docs/prompt-library https://docs.lifesight.io/docs/insights | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The platform documents CSV exports of budget worksheets and CSV/XLSX exports of attribution data. This satisfies the platform-level scoring rule for reusable data in categories 6–8, without implying that equivalent downloadable files are returned directly by MCP. | https://docs.lifesight.io/docs/intrepreting-a-budget-plan https://docs.lifesight.io/docs/overview-1 | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | The ChatGPT/Codex guide documents an external MCP endpoint and token-based authentication. I did not find a repeatable unattended invocation or recurring-workflow example, so external agent automation has partial evidence. | https://docs.lifesight.io/docs/setup-guide-chatgpt | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | MCP answers are described as model-backed, and Lifesight documents a hybrid causal framework using ridge regression, bootstrapping and ensemble forecasting. I did not find evidence that a Bayesian MMM is the core measurement backbone or that identical calls guarantee deterministic answers, so this compound criterion is only partially supported. | https://docs.lifesight.io/docs/mcp-connect https://docs.lifesight.io/docs/lifesights-approach | Technical doc Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Lifesight documents calibrating MMM against incrementality-test results using channel, period, observed return or lift, and confidence inputs. Before-and-after calibration outputs provide strong platform evidence of experiment-informed MMM calibration under the categories 6–8 scoring rule. | https://docs.lifesight.io/docs/model-calibration https://docs.lifesight.io/docs/model-creation | Technical doc Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | Model Attributes documents holdout accuracy, NRMSE, estimation error and actual-versus-predicted validation views. These are explicit model-quality outputs, satisfying this criterion under the platform-level evidence rule for categories 6–8. | https://docs.lifesight.io/docs/model | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 1 | The self-serve model setup exposes adstock, saturation, training split, refresh frequency and causal relationships, while calibration inputs remain inspectable. The modeling framework also describes configurable weakly informative priors, supporting auditable and editable model settings. | https://docs.lifesight.io/docs/model-creation https://docs.lifesight.io/docs/model-calibration https://docs.lifesight.io/docs/lifesights-approach | Technical doc Technical doc Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | I did not find a documented MCP operation allowing authorized agents to propose or apply model-calibration changes with validation and version history. The available calibration instructions describe a platform UI workflow, which does not establish the requested controlled agent interface. | https://docs.lifesight.io/docs/model-calibration https://docs.lifesight.io/docs/mcp-connect | Technical doc Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Lifesight publicly displays numerous enterprise reference logos, including IKEA, Decathlon, Dyson, New Balance, Reckitt, BAT, Mango, San Miguel, Vueling and Bath & Body Works. I did not find revenue evidence on the permitted vendor sources establishing at least ten distinct reference brands above the $1 billion threshold, so the full requirement remains only partially substantiated. | https://lifesight.io/product/optimize/ https://lifesight.io/trust-center/ | Product page Product page |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | The technical SOC 2 page states compliance, and the Trust Center explicitly claims SOC 2 Type II, ISO 27001 and annual independent audits. These vendor-published statements provide strong public evidence of third-party security assurance, although the underlying audit reports were not reviewed. | https://docs.lifesight.io/docs/soc-2 https://lifesight.io/trust-center/ | Technical doc Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 1 | Technical documentation offers workspace data localization in the United States and Europe, and the Trust Center explicitly lists EU and US residency. Access-related data and some summaries are exceptions in the technical documentation, and changing regions requires a new workspace. | https://docs.lifesight.io/docs/data-localization-security-privacy-compliance https://lifesight.io/trust-center/ | Technical doc Product page |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0.5 | The technical localization guide explicitly documents deployment on Google Cloud with regional BigQuery datasets. I did not find a customer deployment choice spanning AWS, GCP and Azure, so the three-cloud requirement receives partial credit. | https://docs.lifesight.io/docs/data-localization-security-privacy-compliance | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | The Trust Center explicitly states support for SSO/SAML and SCIM provisioning. This is strong public platform evidence of enterprise single sign-on. | https://lifesight.io/trust-center/ | Product page |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | I did not find a publicly accessible, promptable MCP sandbox or trial without sales-call gating. The MCP page requires an active workspace and directs prospective users to book a demo for sandbox access. | https://lifesight.io/mcp/ https://docs.lifesight.io/docs/setup-guide | Product page Technical doc |
Measured
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.75 | Measured's MCP returns portfolio performance as an executive summary with cross-channel results and conversion-type detail, and the REST API pulls cross-channel performance data. We did not find a documented report of actual online sales for a specified period. | https://www.measured.com/video/measured-brings-causal-intelligence-into-the-ai-tools-marketers-already-use/ | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.75 | The MCP returns results by conversion type, and the platform models in-store and retail sales, including retailer POS data ingested via Alloy.ai and data from Snowflake. We did not find evidence that the MCP reports physical-store sales from a customer's data warehouse specifically. | https://www.measured.com/video/measured-brings-causal-intelligence-into-the-ai-tools-marketers-already-use/ | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The REST API extracts platform-reported and incrementality metrics spanning all channels, tactics and campaigns into a data warehouse or BI tool, drawing on 300+ data partners. The MCP launch material also cites performance visibility across all major platforms in one place. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | The REST API covers platform-reported metrics for all channels, and the platform measures linear TV, out-of-home, direct mail and (per its integration pages) radio and podcast logs. We did not find offline channels named for the API or MCP, and found no evidence for print. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 0.75 | The MCP returns conversion-type detail, and the model's dependent variable can be orders, new-customer orders, activations or traffic, with CPA reported alongside ROAS. We did not find evidence of contribution margin being reported through the MCP or API. | https://www.measured.com/video/measured-brings-causal-intelligence-into-the-ai-tools-marketers-already-use/ | Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | The REST API returns data at channel, tactic and campaign granularity, and the platform tracks performance by region, brand and retailer. We did not find documented date, brand, market or product filter parameters for the MCP or API. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The REST API pulls incrementality metrics from the Measured dashboard for all channels, tactics and campaigns, and the dashboard reports incremental sales, ROAS and CPA per channel. The MCP returns cross-channel results grounded in Measured's causal methodology. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.75 | The REST API covers incrementality metrics for all channels, and Measured measures linear TV, CTV, out-of-home and direct mail with geo and matched-market tests. We did not find offline channels named explicitly for the API or MCP output. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.25 | The Measured Incrementality Model can include price and promotion indexes as configurable non-media variables with an elasticity interpretation. We found no evidence that promotion-driven revenue is reported as a contribution, via the MCP or elsewhere. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.5 | The Measured Incrementality Model is run daily, with daily results reviewable in its UI, but it operates on weekly data and results reach downstream tools on a client-specified cadence. We did not find evidence that the MCP exposes daily MMM-based incremental ROAS. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.5 | The model has a separate non-media component with intercept, seasonality, holidays, macroeconomic, price, promotion and custom variables, plus a media component giving tactic-level contributions. We did not find evidence that this decomposition is available through the MCP or API. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | The Media Plan Optimizer allocates budgets across channels by marginal ROI to maximize growth or profitability. The MCP provides marginal return data to guide budget allocation, but we did not find evidence that it invokes the optimizer. | https://www.measured.com/media-plan-optimizer/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.5 | The Media Plan Optimizer runs what-if simulations that forecast outcomes for a budget plan before it is committed. We did not find evidence that forecasting is available through the MCP or API. | https://www.measured.com/media-plan-optimizer/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | The MCP launch lists optimization insights with marginal return and incremental growth potential data among its contents. Response curves are generated for every tactic in the platform, but we did not find evidence that the curves themselves are exposed through the MCP. | https://www.measured.com/press/measured-launches-mcp-server/ | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.5 | The Media Plan Optimizer lets users simulate budget changes by channel and conversion type and see projected incremental outcomes. We did not find evidence that this simulation can be run through the MCP or API. | https://www.measured.com/media-plan-optimizer/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.5 | In the Media Plan Optimizer, users set a target budget, reference period, optimization strength and budget limits for any tactic. We did not find evidence that constraints can be passed through the MCP or API. | https://www.measured.com/blog/navigate-the-maze-of-budget-planning-with-media-plan-optimizer/ | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.5 | The Media Plan Optimizer starts from a predefined baseline plan and lets users compare several what-if plans side by side. We did not find evidence that scenario comparison is available through the MCP or API. | https://www.measured.com/blog/smarter-omnichannel-marketing-starts-now-introducing-measureds-media-plan-optimizer/ | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.5 | The Media Plan Optimizer plans toward revenue, ROAS, new-customer or profitability goals. We did not find evidence that goal-based planning is available through the MCP or API. | https://www.measured.com/media-plan-optimizer/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0.5 | The platform keeps saved media plans, which include recommended spend levels for every campaign and can be shared with teams. We did not find evidence that saved plans can be retrieved through the MCP or API. | https://www.measured.com/blog/product-update-incrementality-powered-media-planning-and-new-benchmarks-categories/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The REST API extracts incrementality metrics down to campaign level, and Measured's Facebook testing measures incrementality at ad set level. We did not find ad set level results documented for the API or MCP. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | The REST API is described as giving a side-by-side view of platform-reported and incrementality metrics across channels, tactics and campaigns. We did not find last-click ROAS included in that comparison. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.25 | Response curves, and so marginal returns, are modeled at the tactic level in the Measured Incrementality Model. We did not find evidence of marginal returns for individual campaigns or ad sets. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | Saved Media Plan Optimizer plans include recommended spend levels for every campaign. We did not find evidence of daily budget recommendations, or of campaign budget recommendations through the MCP or API. | https://www.measured.com/blog/product-update-incrementality-powered-media-planning-and-new-benchmarks-categories/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0 | We did not find evidence that Measured recommends bid values such as target ROAS for campaigns or ad sets. Its commentary on Google's target-based bidding changes is advisory and does not describe a product feature. | https://www.measured.com/blog/google-ads-is-changing-target-based-bidding-what-advertisers-need-to-know/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0 | We did not find evidence that the MCP or platform pushes budget changes to Meta, Google or other ad platform APIs. Optimized plans are pushed into internal tools and dashboards, and test campaigns are trafficked automatically, but budget changes are not described. | https://www.measured.com/blog/smarter-omnichannel-marketing-starts-now-introducing-measureds-media-plan-optimizer/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | We did not find evidence that the MCP or platform pushes bidding changes such as target ROAS to ad platform APIs. The planning material only describes sending plans to channel managers or internal tools. | https://www.measured.com/blog/smarter-omnichannel-marketing-starts-now-introducing-measureds-media-plan-optimizer/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | The Optimization Report compares the prior period's budget allocation with actual performance by tactic to show revenue gained from budget changes. We did not find campaign or ad set level pre/post analysis of bidding changes, or evidence of this via the MCP. | https://www.measured.com/faq/what-is-the-optimization-report/ | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | The MCP draws on an incrementality database of recent tests, and the MCP demo grounds test recommendations in the customer's own results, such as a stale Google non-brand search holdout. We did not find explicit documentation of retrieving individual geo test results with intervention and outcome through the MCP. | https://www.measured.com/press/measured-launches-mcp-server/ | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.5 | Measured runs audience-split tests for owned channels such as direct mail, catalog, email and SMS, as in the Outerknown direct mail case. We did not find evidence of this via the MCP, and the DPA describes the Known Audience Testing service as legacy. | https://www.measured.com/triangulated-measurement/ | Product page |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.25 | The model is calibrated mainly with geo holdout tests, and clients can add CPO priors from non-Measured experiments. We did not find evidence that Meta Conversion Lift results are retrieved or reported, and Measured positions platform lift studies as directional input only. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 1 | An MCP demo shows the AI returning a prioritized testing plan with near-term and longer-term experiments, grounded in the customer's results and priors. Examples include a Meta prospecting scale test and a revalidation of a stale Google non-brand search holdout. | https://www.measured.com/video/ask-measured-what-to-test-next-right-inside-your-ai/ | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | The Geo Designer lets users set test length, holdout size, budget and markets while showing in real time the minimum detectable effect, revenue risk and likelihood of detecting lift. We did not find evidence that test design is available through the MCP. | https://www.measured.com/press/measured-evolves-geo-testing-experience/ | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 0.5 | Measured states that its MCP server works inside ChatGPT, Claude, Gemini and other AI platforms. We did not find public connection or authentication instructions for Claude or ChatGPT. | https://www.measured.com/press/measured-launches-mcp-server/ | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0.5 | A customer quoted in the launch says MCP responses come back already formatted, and a demo shows a recommendation published as a presentation-ready deck. We did not find evidence that the server itself returns table or chart artifacts rather than the AI client rendering them. | https://www.measured.com/press/measured-launches-mcp-server/ | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0 | We did not find public documentation of the MCP tool definitions or a metric reference for interpreting its output. The launch pages describe capabilities in general terms and direct readers to request a demo. | https://www.measured.com/video/measured-launches-mcp-server-giving-enterprise-marketers-on-demand-access-to-incrementality-data-inside-the-ai-tools-they-already-use/ | Marketing collateral |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The REST API pulls cross-channel platform-reported and incrementality metrics into a customer's database, data warehouse or BI tools. The platform also exports reports as files and pushes data extracts to Amazon S3, Azure or Google Cloud storage. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | The REST API is documented for continuously pulling cross-channel data into a data warehouse in near real time. We did not find documentation of repeatable MCP use from an external agent runtime, and AI-powered workflows and predefined skills are listed as future capabilities. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | The Measured Incrementality Model is a regression-based MMM with Bayesian-style priors, but Measured states it is not fully Bayesian and is fitted by customized maximum likelihood rather than MCMC. The MCP's answers are described as grounded in this causal methodology. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Geo holdout results enter the model as priors weighted by recency and confidence, with no limit on the number of studies, and coefficients are built to align closely with them. Clients can also add CPO priors from other experiments or in-house MMMs. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | A daily back-test computes out-of-sample NRMSE and MAPE over 8, 12 and 16 week holdouts, and each tactic is evaluated on p-value, confidence interval width, VIF, sufficiency and alignment. Daily model results are available for review in the Measured Incrementality Model UI. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 1 | Users configure non-media variables, include or exclude geo study results and select tactic inputs in the UI, and an admin interface shows which variables and test priors drive outputs. Settings not yet in the UI can be changed on request through customer success. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We did not find evidence that agents can propose or apply model calibration changes through the MCP or API. The MCP is described as giving read access to insights, while calibration is configured in the platform UI. | https://www.measured.com/press/measured-launches-mcp-server/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Named customers on Measured's site include The North Face, VF Corp, Unilever, Patagonia, Crocs and Overstock, which are clearly $1B+ revenue brands. Most other named references (e.g. Vuori, Dermalogica, Faherty, Johnny Was, Otter.ai) are smaller or uncertain, so we found fewer than 10 at $1B+. | https://www.measured.com/customers/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Measured states it is ISO/IEC 27001:2022 certified and SOC 2 Type 2 compliant, with Sprinto used for continuous compliance monitoring. The certificate, SOC 2 report and penetration test results are shared under NDA. | https://www.measured.com/security-compliance/ | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | We did not find evidence of a choice between US and EU data residency. Measured states that it stores all client data within the United States, with EU transfers covered by Standard Contractual Clauses. | https://www.measured.com/security-compliance/ | Product page |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We did not find evidence of a choice between AWS, GCP and Azure for hosting. Amazon Web Services is the only infrastructure provider in the DPA's subprocessor list, and Azure and Google Cloud appear only as destinations for customer data exports. | https://www.measured.com/media/Measured-DPA-03-06-2026.pdf | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | We did not find evidence of enterprise SSO (SAML or federation with a customer's identity provider). Measured states that platform login is powered by Okta with mandatory multi-factor authentication, and the DPA refers to username and password logins. | https://www.measured.com/data-protection/ | Product page |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We did not find a public, promptable MCP demo, trial or sandbox. The MCP pages show recorded demo videos and direct visitors to request a demo. | https://www.measured.com/video/measured-launches-mcp-server-giving-enterprise-marketers-on-demand-access-to-incrementality-data-inside-the-ai-tools-they-already-use/ | Marketing collateral |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.75 | The REST API exposes dashboard performance data, and the platform reconciles sales with media spend. Evidence is partial because a dated actual-online-sales response is not explicitly documented. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/press/measured-secures-funding-to-solve-attribution-for-advertisers/ | Product page Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.75 | API reporting and omnichannel measurement together support access to retail sales-related results. I did not find an explicit MCP or API example returning actual store sales sourced from a customer warehouse for specified dates. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/omnichannel-measurement/ | Product page Product page |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The REST API exports platform-reported cross-channel metrics through the dashboard. Together with documented media-spend reporting and managed integrations, this strongly supports programmatic digital-media reporting under the API-equivalence rule. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/cross-channel-dashboard/ https://www.measured.com/integrations/ | Product page Product page Product page |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | The REST API covers all dashboard channels, and Measured documents offline media coverage. Evidence is partial because I did not find an explicit reporting specification covering spend and metrics for TV, out-of-home, radio and print together. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/media-mix-modeling/ | Product page Product page |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | MCP examples report Total Orders. This directly satisfies the criterion through a documented business outcome beyond revenue. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | MCP examples select brands and comparison periods, while the API supports campaign-level data. I did not find complete documentation for explicit market and product filters or all requested grouping dimensions. | https://support.measured.com/article/elll7zw680-connect-measured-mcp-to-claude-ai https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://www.measured.com/integration/measured-rest-api/ | Technical doc Technical doc Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | MCP examples retrieve channel iROAS and incremental sales. This is direct evidence of causal performance reporting rather than platform attribution alone. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.75 | Programmatic dashboard exports include incremental metrics across channels, and the model covers offline media. I did not find an MCP or API example explicitly enumerating offline-channel revenue and iROAS outputs for TV, out-of-home and radio. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/media-mix-modeling/ | Product page Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.75 | MCP exposes non-media decomposition, and MIM includes configurable price and promotion effects. Evidence is partial because a separate promotion-driven revenue output is not explicitly demonstrated. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.75 | MCP exposes MIM results, whose daily runs are available in the model UI. Daily MCP freshness remains unconfirmed because downstream publication follows a client-selected cadence and the product page also describes weekly insights. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://www.measured.com/faq/measured-incrementality-model-explainer/ https://www.measured.com/media-mix-modeling/ | Technical doc Technical doc Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | MCP examples expose media/non-media composition and negative non-media drivers. Together with the documented model baseline, this strongly supports performance-driver decomposition. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | The platform optimizer calculates channel allocations. I found no evidence that MCP or the public REST API invokes a new optimization run for a supplied objective and period. | https://www.measured.com/media-plan-optimizer/ https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Product page Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.5 | The platform forecasts customizable media plans. I found no documented MCP or API operation that invokes forecasting for a newly specified allocation. | https://www.measured.com/media-plan-optimizer/ https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Product page Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | The MCP launch explicitly includes marginal-return insights. Response curves are documented in the platform, but I did not find evidence of MCP or API retrieval of the curve data itself. | https://www.measured.com/press/measured-launches-mcp-server/ https://www.measured.com/media-plan-optimizer/ | Marketing collateral Product page |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.5 | The platform runs what-if budget simulations. I found no evidence that MCP or the public REST API calculates a fresh scenario from a user-specified spend change. | https://www.measured.com/media-plan-optimizer/ https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Product page Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.5 | The optimizer supports tactic budget limits and planning periods. Reading saved constraints through MCP does not establish an operation that passes and enforces new user-defined constraints. | https://www.measured.com/blog/navigate-the-maze-of-budget-planning-with-media-plan-optimizer/ https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Marketing collateral Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.75 | MCP compares saved plans on budget, returns and constraints. Baseline planning is documented in the platform, but an explicit MCP baseline-versus-alternative comparison is not demonstrated. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://www.measured.com/blog/smarter-omnichannel-marketing-starts-now-introducing-measureds-media-plan-optimizer/ | Technical doc Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.5 | The platform builds plans toward revenue, ROAS and profitability goals. I found no evidence of an MCP or API operation calculating a new plan against a supplied business target. | https://www.measured.com/media-plan-optimizer/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0.75 | MCP lists saved plans and compares their inputs and outcomes. Evidence is partial because stable plan identifiers and a complete retrievable assumptions schema are not documented. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The REST API exports campaign-level incrementality metrics, and broader platform material names ad-set measurement. I did not find explicit programmatic coverage of both incremental revenue and iROAS for individual ad sets. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/press/measured-secures-funding-to-solve-attribution-for-advertisers/ | Product page Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | The REST API provides campaign-level platform and incremental metrics side by side. I did not find an explicit three-way comparison including last-click ROAS at both campaign and ad-set level. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.5 | Measured documents ad-set response curves and marginal metrics in its scale-testing solution. I found no evidence of MCP or API retrieval of marginal returns at both campaign and ad-set level. | https://www.measured.com/faq/scale-facebook-campaigns/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | Platform material describes granular budget-allocation recommendations for campaigns and ad sets. I found no evidence that these are optimal daily budgets, or that MCP or an API returns them at both levels. | https://www.measured.com/press/measured-secures-funding-to-solve-attribution-for-advertisers/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0 | I found no evidence of per-campaign or per-ad-set optimal bid-target recommendations through MCP, API or the broader platform. The reviewed bidding article gives advertisers guidance rather than documenting a Measured bid-recommendation feature. | https://www.measured.com/blog/google-ads-is-changing-target-based-bidding-what-advertisers-need-to-know/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.25 | Measured documents automatic test-campaign trafficking and scale-spend recommendations, providing limited platform evidence of campaign activation. I found no documented MCP or API operation for routine daily budget changes on Meta or Google. | https://www.measured.com/blog/enhanced-automation-easy-campaign-mapping-and-deeper-geo-testing-customization/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | I found no evidence that MCP, a Measured API or the broader platform pushes bid parameters to advertising platforms. The reviewed material documents measurement access and bidding advice rather than a bid-write operation. | https://www.measured.com/integration/measured-rest-api/ https://www.measured.com/blog/google-ads-is-changing-target-based-bidding-what-advertisers-need-to-know/ | Product page Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0 | I found no evidence of a report tied to each executed campaign or ad-set bidding change that compares actual pre/post revenue and spend. Geo-test result retrieval does not establish that specific change-level workflow. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://www.measured.com/blog/google-ads-is-changing-target-based-bidding-what-advertisers-need-to-know/ | Technical doc Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 1 | MCP documents geo-test summaries and weekly result breakdowns. This directly supports retrieval of completed incrementality-test findings. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.5 | The Outerknown case documents controlled direct-mail audience tests and incremental profitability results. I found no evidence that MCP or the public API retrieves the underlying owned-media A/B test results. | https://measured.com/media/Case-Study-Outerknown.pdf | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0 | I found no evidence that MCP, the public API or the broader platform retrieves Meta Conversion Lift study results. The reviewed Meta material explains lift methods and Measured triangulation without documenting ingestion or retrieval of those studies. | https://www.measured.com/faq/can-i-measure-incrementality-on-facebook/ | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 1 | Measured demonstrates an AI-based prioritized testing plan grounded in prior results. The examples identify Meta scale headroom and stale Google holdout evidence, directly supporting targeted experiment recommendations. | https://www.measured.com/video/ask-measured-what-to-test-next-right-inside-your-ai/ | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | Measured documents automated geo-test design, market selection and recommended scale spend, alongside guidance on duration and statistical power. I found no MCP or API operation producing the complete treatment/control, power, duration and spend design. | https://www.measured.com/incrementality-testing/ https://www.measured.com/blog/enhanced-automation-easy-campaign-mapping-and-deeper-geo-testing-customization/ https://www.measured.com/faq/how-to-run-geo-testing-for-marketers-a-step-by-step-guide/ | Product page Marketing collateral Marketing collateral |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Public technical guides provide setup and OAuth authentication instructions for both ChatGPT and Claude. These include the server address, organizational setup and individual account connection. | https://support.measured.com/article/q44g6jepr3-connect-measured-mcp-to-chat-gpt https://support.measured.com/article/elll7zw680-connect-measured-mcp-to-claude-ai | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0.5 | MCP examples request tables and formatted reports. I did not find confirmation that the server itself returns native chart or table artifacts rather than the client formatting returned data. | https://support.measured.com/article/9uvsd8qfax-measured-mcp-prompt-library https://support.measured.com/article/elll7zw680-connect-measured-mcp-to-claude-ai | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0.5 | Public guidance explains MCP workflows, and Measured publishes metric and model methodology. I did not find a public catalog of concrete tool names, parameter schemas and result definitions. | https://measured.mintlify.app/connect-mcp-clients https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The REST API programmatically exports platform and incremental metrics into databases, warehouses and BI tools. This is strong evidence of reusable structured data under the API-equivalence rule. | https://www.measured.com/integration/measured-rest-api/ | Product page |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | Measured documents connecting its MCP tools to Copilot Studio agents. I did not find a repeatable invocation example or recurring-analysis workflow, so automation evidence remains partial. | https://measured.mintlify.app/mcp-clients/copilot | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | MIM uses constrained regression and Bayesian-style priors with maximum-likelihood fitting. This supports model-backed answers, but the methodology explicitly says the model is not fully Bayesian. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | MIM ingests geo-experiment contributions as weighted causal priors. This directly establishes incrementality-test calibration of the underlying model. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | Measured explicitly reports model fit, prediction error and out-of-sample validation. Its methodology also describes daily holdout back-testing, supporting the platform-level requirement for model-quality reporting. | https://www.measured.com/faq/qa-media-mix-modeling-mmm/ https://www.measured.com/faq/measured-incrementality-model-explainer/ | Marketing collateral Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | MIM exposes several settings in its UI, including geo-study inclusion and tactic input selection. Some configuration still requires customer-success assistance, and I found no complete self-serve specification for inspecting and editing priors. | https://www.measured.com/faq/measured-incrementality-model-explainer/ | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | I found no evidence of MCP or API operations that propose or apply calibration changes with validation and version history. The reviewed methodology describes UI configuration and customer-success requests, which do not establish the required agent workflow. | https://www.measured.com/faq/measured-incrementality-model-explainer/ https://measured.mintlify.app/connect-mcp-clients | Technical doc Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Measured publicly identifies enterprise customers, including The North Face, and describes multinational deployments. I could not verify at least ten named reference brands each exceeding $1 billion in revenue using the permitted sources. | https://www.measured.com/customers/ https://www.measured.com/enterprise/ | Product page Product page |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Measured states ISO/IEC 27001:2022 certification and SOC 2 Type 2 compliance. It offers certificates and audit reports under NDA, providing strong public evidence of independent security assurance. | https://www.measured.com/security-compliance/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | Measured explicitly states that all client data is stored in the United States. I found no evidence of a customer-selectable EU data-residency option. | https://www.measured.com/security-compliance/ | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | I found no evidence that customers can choose deployment on AWS, GCP or Azure. The reviewed infrastructure and integration pages do not establish a multi-cloud hosting choice. | https://www.measured.com/enterprise/ https://www.measured.com/integrations/ | Product page Product page |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | The MCP authentication guide specifies Okta SSO using existing Measured portal credentials. This directly supports enterprise single sign-on. | https://support.measured.com/article/elll7zw680-connect-measured-mcp-to-claude-ai | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | Public MCP setup requires a Measured account with access enabled. I found demonstration videos but no ungated, hands-on promptable MCP trial. | https://measured.mintlify.app/connect-mcp-clients https://www.measured.com/video/measured-brings-causal-intelligence-into-the-ai-tools-marketers-already-use/ | Technical doc Marketing collateral |
Northbeam
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | Northbeam MCP reads live dashboard data including revenue and orders, with documented prompts such as spend and revenue by platform for the last 30 days and top orders by revenue this month. The Data Export API adds revenue, transactions and orders for any fixed or relative period at daily, weekly or monthly granularity. | https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.25 | We did not find evidence that the MCP reports physical-store sales, and the Revenue Source breakdown documents only Online Store and Amazon groups. The Orders API accepts orders not placed on the live website (tagged as offline orders), and MMM+ marketing mentions retail revenue, but we found no store or data warehouse ingestion described. | https://docs.northbeam.io/docs/orders-api | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The MCP FAQ says the assistant can access performance metrics and attribution data, and the documented prompts cover Meta ROAS and spend by platform. Spend, impressions, CPM and CTR are pulled via API from Meta, Google, TikTok, Microsoft, Snapchat and other platforms and are exposed through the Data Export API at platform, campaign, ad set and ad level. | https://docs.northbeam.io/docs/northbeam-mcp-server-faqs | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | Offline and non-integrated media spend, clicks and impressions can be loaded through the Spend API or a spend sheet, and TV and audio arrive through Tatari (streaming, with linear on request) and Bliss Point integrations; this spend then appears in the same platform-level data the MCP reads. We did not find native or documented support for out-of-home, radio or print data. | https://docs.northbeam.io/docs/spend-api | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The Data Export API and dashboards expose orders, new and returning customers, CAC, AOV, MER, subscription transactions and LTV metrics, and the MCP reads the same dashboard metrics and orders. We did not find a contribution margin metric; Profit Benchmarks use COGS only to derive ROAS and CAC targets. | https://docs.northbeam.io/docs/northbeam-api-data-export-1 | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | The Data Export API supports explicit date periods, platform, campaign, ad set and ad levels, breakdown filters and hourly to monthly granularity, and the MCP lets users choose which dashboard to query. Product filtering is Shopify-only, and brand or market separation requires separate dashboards or order segmentation, which does not segment spend. | https://docs.northbeam.io/docs/northbeam-api-data-export-1 | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 0.5 | We did not find evidence that the MCP or API returns incremental ROAS; the MCP exposes multi-touch attribution metrics, and the export API lists only attribution models such as Clicks Only and Last Touch. The broader platform measures channel-level incrementality through MMM+ and reports iROAS and iCAC from geo incrementality tests. | https://docs.northbeam.io/docs/incrementality | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.25 | We did not find evidence that the MCP reports incremental ROAS for offline channels. MMM+ is described as measuring media that is hard to measure on a click basis at channel level, but the public documentation does not show offline channel iROAS outputs. | https://docs.northbeam.io/docs/mixed-media-modeling | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.5 | We did not find evidence that promotion-driven revenue is available through the MCP or API. MMM+ models promotional periods and seasonality, and the product page highlights promotional and sales sensitivity and spend guidance during promos. | https://www.northbeam.io/products/mmm-plus | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.25 | We did not find evidence that MMM-based incremental ROAS is available through the MCP. The MMM+ product page claims daily optimizations and model training, but a Northbeam blog describes weekly retraining and the documentation says MMM updates more slowly than attribution. | https://www.northbeam.io/products/mmm-plus | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.25 | We did not find evidence that the MCP exposes modeled contributions. The MMM documentation describes time-series regression controlling for promotions and seasonality, but we found no documented base, media and non-media decomposition output. | https://docs.northbeam.io/docs/mixed-media-modeling | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | We did not find evidence that the MCP invokes an optimizer; the MCP is documented as read-only. MMM+ is marketed as recommending the optimal omnichannel budget allocation with daily recommendations and optimizations. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.5 | We did not find evidence that forecasting is exposed through the MCP or API. MMM+ is marketed with flexible forecasts, live daily performance forecasts and budget scenarios. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.25 | We did not find evidence that marginal returns or response curves are available through the MCP. The MMM+ product page only refers to avoiding diminishing returns, without documenting response curve or marginal ROAS outputs. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.5 | We did not find evidence that budget simulations can be run through the MCP. Northbeam's MMM demo page describes a self-service MMM dashboard where users adjust budget mixes on the fly, and the product page mentions spend scenarios. | https://www.northbeam.io/mmm-demo | Marketing collateral |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0 | We did not find evidence that the MCP or the broader platform supports user-defined budget limits or channel restrictions in planning. The public MMM+ material does not describe optimizer constraints. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.25 | We did not find evidence that scenario comparison is available through the MCP. MMM+ mentions ideal spend scenarios and budget scenarios, but we found no documentation of comparison against an explicit baseline plan. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.25 | We did not find evidence that the MCP plans against a business target. Profit Benchmarks derive account-wide ROAS, MER and CAC targets from profitability inputs such as COGS, but these are targets rather than a budget plan for a specified outcome. | https://docs.northbeam.io/docs/profitability-benchmarks | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0 | We did not find evidence that saved scenarios can be retrieved through the MCP or that the platform documents saved plans with identifiers. The MMM+ material mentions scenarios but not saving or retrieving them. | https://www.northbeam.io/products/mmm-plus | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.25 | We did not find evidence that the MCP reports incremental ROAS by campaign or ad set, and the MMM documentation states it reports at channel and business level, not campaign or ad level. Incrementality tests can target a campaign or channel, which is partial platform evidence. | https://docs.northbeam.io/docs/mixed-media-modeling | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | Through the MCP and Data Export API, campaign and ad set results are available under several attribution models including Last Touch and Last Non-Direct Touch, alongside ad-platform reported conversions pulled from each platform. We did not find incremental ROAS at campaign or ad set level to complete the three-way comparison. | https://docs.northbeam.io/docs/northbeam-api-data-export-1 | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0 | We did not find evidence of marginal returns for individual campaigns or ad sets through the MCP or in the broader platform. The MMM documentation describes channel-level reporting only. | https://docs.northbeam.io/docs/mixed-media-modeling | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | We did not find evidence that the MCP returns daily budget recommendations for campaigns or ad sets. MMM+ is marketed with daily recommendations and spend scenarios, but at channel level. | https://www.northbeam.io/products/mmm-plus | Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | We did not find evidence that the MCP returns bidding targets. Profit Benchmarks provide account-wide ROAS, MER and CAC targets shown as indicators on campaigns and ads, which is partial evidence rather than per-campaign bid values. | https://docs.northbeam.io/docs/profitability-benchmarks | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0 | We did not find evidence that the MCP or the platform pushes budget changes to ad platforms; the MCP is documented as read-only. Northbeam Apex sends attribution data to Meta for delivery optimization but is not described as changing budgets. | https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | We did not find evidence that the MCP or the platform pushes bid or Target ROAS changes to ad platforms. Apex feeds Northbeam-attributed conversions into Meta Custom Attribution, which users configure in Meta Ads Manager, rather than changing bid parameters. | https://docs.northbeam.io/docs/northbeam-apex | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0 | We did not find evidence of pre/post analysis of executed bidding or budget changes through the MCP or in the platform. The closest feature, Metrics Explorer, is a correlation tool that the documentation says does not imply causation. | https://docs.northbeam.io/docs/metrics-explorer | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.5 | We did not find evidence that incrementality test results are available through the MCP or API. The platform runs geo incrementality tests and reports lift, iROAS, iCAC, confidence intervals and a significance verdict for the tested channel. | https://docs.northbeam.io/docs/incrementality | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0 | We did not find evidence of owned-media A/B or holdout tests such as email or leaflet experiments, either through the MCP or in the platform. The incrementality documentation focuses on paid media channels and geo testing. | https://docs.northbeam.io/docs/incrementality | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0 | We did not find evidence that the MCP or the platform ingests or reports Meta Conversion Lift results. The Apex FAQ mentions Meta Incremental Attribution only as an ad set setting that cannot be combined with Custom Attribution. | https://docs.northbeam.io/docs/apex-faqs | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | We did not find evidence that the MCP prioritizes experiments. The documentation offers general guidance to test high-spend channels with uncertain attribution, and the product page promises recommendations after each test, but no prioritization engine is described. | https://docs.northbeam.io/docs/use-cases/validate-with-incrementality | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | We did not find evidence that test design is available through the MCP. The incrementality product page says it generates experiment designs based on spend, channel mix, conversion lag and KPIs, and the documentation requires sufficient statistical power, though no explicit MDE or duration output is documented. | https://www.northbeam.io/products/incrementality | Product page |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Northbeam publishes step-by-step guides for Claude (web, Desktop and Claude Code, including Team and Enterprise admin setup) and ChatGPT (developer mode and workspace setup) using the https://mcp.northbeam.io server. Authentication uses the normal Northbeam login via a browser sign-in flow. | https://docs.northbeam.io/docs/connect-northbeam-to-claude | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | We did not find evidence that the MCP returns server-provided tables, charts or other visual artifacts. The documentation describes the assistant querying and summarizing data only. | https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0.5 | The Data Export API has discovery endpoints for metrics, breakdowns and attribution models, and Metrics 101 documents metric definitions. We did not find public documentation of the MCP tool definitions; the MCP pages only give example prompts. | https://docs.northbeam.io/docs/northbeam-metrics-101 | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The Data Export API returns complete CSV exports via temporary link or to the customer's GCS or S3 bucket, and Sales page exports can be scheduled daily, weekly or monthly. The documentation notes the MCP itself retrieves limited data per question and is not a bulk export. | https://docs.northbeam.io/docs/exporting-data | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | Northbeam documents MCP setup in Claude Code and in any AI tool that supports custom connectors, and the Data Export API is positioned for automated pipelines with API keys. We did not find documentation of recurring or scheduled MCP invocation from an external agent runtime. | https://docs.northbeam.io/docs/connect-northbeam-to-ai-agent | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0 | We did not find evidence that MCP answers are grounded in a Bayesian MMM; the MCP exposes multi-touch attribution metrics. Northbeam describes MMM+ as machine learning and time-series regression, without mention of a Bayesian approach. | https://docs.northbeam.io/docs/mixed-media-modeling | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 0.5 | The incrementality documentation says test results feed into MMM to replace correlation-based estimates with causal ones, so the platform MMM is calibrated with tests. We did not find evidence that the MCP uses the MMM, so the calibration does not clearly carry over to MCP answers. | https://docs.northbeam.io/docs/incrementality | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 0 | We did not find evidence that the platform or MCP reports model validation metrics such as R2, MAPE or holdout performance. The MMM documentation only notes that quality depends on enough history and clean data. | https://docs.northbeam.io/docs/mixed-media-modeling | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | Northbeam's MMM demo page describes a self-service MMM dashboard where users can adjust models and budget mixes. We did not find evidence that priors or other calibration settings can be inspected or edited. | https://www.northbeam.io/mmm-demo | Marketing collateral |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We did not find evidence of model calibration through the MCP, which is documented as read-only and unable to change anything in Northbeam. Calibration with incrementality results is described only as a platform process. | https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0 | We did not find evidence of 10 public reference customers with $1B+ revenue on Northbeam's website. The customer stories page names about 18 brands, mostly DTC (for example Timex, HexClad, Dr. Squatch, PetMeds), and gives no revenue figures. | https://www.northbeam.io/customer-stories | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Northbeam's data security page states that it is SOC 2 Type 2 compliant and that the report is available on request. We did not find evidence of ISO 27001 certification. | https://www.northbeam.io/data-security | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | We did not find evidence of a choice between US and EU data residency. The privacy policy states services are hosted and operated in the United States, and EU transfers rely on Standard Contractual Clauses. | https://www.northbeam.io/privacy | Product page |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We did not find evidence that customers can choose between AWS, GCP and Azure. Northbeam runs most of its platform in Google Cloud, with AWS listed as a subprocessor, and exports can be sent to the customer's own GCS or S3 bucket. | https://www.northbeam.io/data-security | Product page |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | We did not find evidence that Northbeam supports customer SSO such as SAML or Okta. User documentation describes email invitations with passwords, and Auth0 is listed only as Northbeam's own identity provider. | https://docs.northbeam.io/docs/adding-or-removing-users-copy | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We did not find evidence of a trial or public demo of the MCP without a sales call. The MCP is available only to Pro and Enterprise customers, and every pricing tier routes to booking a demo or requesting a quote. | https://www.northbeam.io/pricing | Product page |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The Orders Export API returns order revenue for an explicit start and end date, and MCP can retrieve orders by revenue and period. This directly supports reporting actual online sales through the permitted MCP/API routes. | https://docs.northbeam.io/docs/northbeam-api-orders-export https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.75 | The Orders API accepts tagged offline orders from customer systems, and the export API returns dated order records with source names and tags. This is partial API evidence for store-sales reporting, but I did not find an explicit physical-store warehouse retrieval workflow. | https://docs.northbeam.io/docs/orders-api.md https://docs.northbeam.io/docs/northbeam-api-orders-export | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The Data Export API exposes spend and performance metrics from connected Meta, Google, and TikTok accounts. MCP also documents spend and revenue breakdowns by platform for a requested period. | https://docs.northbeam.io/docs/northbeam-api-data-export-1 https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | The Spend API can list customer-supplied spend for non-integrated channels. I found only partial API evidence for the required offline coverage because TV, out-of-home, radio, and print are not all explicitly documented and the setup guide says custom impression and click fields are ignored. | https://docs.northbeam.io/docs/spend-api.md https://docs.northbeam.io/docs/non-integrated-channel-setup-spend-api-utms.md | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The Orders Export API includes orders and first-time versus returning customer classifications. The Data Export API additionally exposes transactions and acquisition metrics, establishing business-outcome reporting beyond revenue. | https://docs.northbeam.io/docs/northbeam-api-orders-export https://docs.northbeam.io/docs/northbeam-api-data-export-1 | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | The export API supports selectable reporting levels, time granularity, and breakdown filters, while MCP documents product-level attribution when enabled. This is partial MCP/API evidence because I did not find complete support for every requested brand, market, and product filter, and order-segment exports are explicitly unsupported. | https://docs.northbeam.io/docs/northbeam-api-data-export-1 https://docs.northbeam.io/docs/northbeam-mcp-server-faqs https://docs.northbeam.io/docs/order-segmentation | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 0.5 | Northbeam documents channel-test readouts with incremental revenue and iROAS, providing strong platform evidence for causal channel measurement. I did not find a documented MCP or API tool retrieving those causal results for digital channels. | https://docs.northbeam.io/interpreting-your-results.md https://docs.northbeam.io/docs/incrementality.md | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.25 | Northbeam describes incrementality testing that can quantify halo effects across retail and offline sales. This is partial platform evidence because I did not find incremental revenue and iROAS readouts for each offline media channel or MCP/API access to them. | https://www.northbeam.io/blog/introducing-incrementality-by-northbeam-automated-lift-testing-you-can-trust | Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 0.25 | The pricing page states that MMM+ measures promotional impact. This supports partial platform credit, but I did not find a promotion-driven revenue decomposition or documented MCP/API retrieval. | https://www.northbeam.io/pricing | Product page |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.25 | The MMM+ product page advertises daily model training and forecasts, while its MMM guide describes weekly retraining. This conflicting platform evidence warrants partial credit because I did not find confirmation of daily MMM-based iROAS measurement or its MCP/API availability. | https://www.northbeam.io/products/mmm-plus https://www.northbeam.io/blog/media-mix-modeling-mmm-guide | Product page Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.25 | The MMM technical guide describes channel contributions and controls for promotions and seasonality. This is partial platform evidence because I did not find an exposed decomposition into baseline, media, and non-media contributions or an MCP/API equivalent. | https://docs.northbeam.io/docs/mixed-media-modeling.md | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.5 | MMM+ explicitly provides budget-allocation recommendations and optimized spend scenarios. I did not find a documented MCP/API operation invoking that optimization engine for a specified objective and period. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.5 | Northbeam describes revenue forecasting from a chosen marketing mix and flexible budget scenarios. I did not find public documentation for invoking those forecasts through MCP or API. | https://www.northbeam.io/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.25 | MMM+ advertises guidance for avoiding diminishing returns, supporting partial platform evidence for spend-response analysis. I did not find an explicit marginal-return and response-curve output specification or MCP/API retrieval route. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.5 | The MMM+ demo page describes self-service adjustment of models and budget mixes, alongside revenue forecasting. I did not find an MCP/API operation that simulates a specified channel-budget change. | https://www.northbeam.io/mmm-demo | Product page |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0 | I did not find evidence that the optimizer accepts and enforces user-defined channel restrictions, budget bounds, and planning dates through MCP/API or the platform UI. The reviewed MMM+ page describes budget optimization without documenting those constraint controls. | https://www.northbeam.io/products/mmm-plus | Product page |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.25 | Northbeam documents flexible budget scenarios and corresponding revenue forecasts. This is partial platform evidence because I did not find an explicit baseline-versus-alternative comparison output or MCP/API access. | https://www.northbeam.io/ | Product page |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.25 | MMM+ promotes ROI-oriented budget optimization, while Profit Benchmarks derives performance targets from profitability goals. These provide partial platform evidence, but I did not find a target-solving budget planner or its MCP/API invocation. | https://www.northbeam.io/products/mmm-plus https://docs.northbeam.io/docs/profitability-benchmarks.md | Product page Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 0 | I did not find evidence of retrieving saved planning scenarios with identifiers, assumptions, and forecast outputs through MCP/API or the broader platform. The documented saved views store reporting configurations rather than modeled budget plans. | https://docs.northbeam.io/docs/manage-breakdowns | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.25 | The incrementality guide describes causal testing of a channel or campaign, providing partial platform evidence for campaign-level measurement. I did not find results for each campaign and ad set or their MCP/API exposure, and MMM documentation limits its granularity to channels and the business. | https://docs.northbeam.io/docs/incrementality.md https://docs.northbeam.io/docs/mixed-media-modeling.md | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | The export API supports multiple attribution models and connected ad-platform metrics at granular reporting levels. This supplies partial API evidence for return comparisons, but I did not find causal incremental ROAS alongside last-click and platform ROAS for each campaign and ad set. | https://docs.northbeam.io/docs/northbeam-api-data-export-1 https://docs.northbeam.io/docs/what-is-northbeam-model-comparison-tool | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0 | I did not find evidence of modeled marginal incremental returns for individual campaigns and ad sets through MCP/API or the broader platform. Northbeam explicitly describes MMM reporting as channel and business level, with MTA used for finer granularity. | https://docs.northbeam.io/docs/mixed-media-modeling.md | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0 | I did not find evidence of optimal daily budget amounts for each campaign and ad set through MCP/API or the platform. Profit Benchmarks supplies profitability targets and performance indicators, without documenting daily spend recommendations at those levels. | https://docs.northbeam.io/docs/profitability-benchmarks.md | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | Profit Benchmarks calculates ROAS and other performance targets that users can apply when assessing campaigns. This is partial platform evidence because I did not find optimal bid-value recommendations for each campaign and ad set or an MCP/API route. | https://docs.northbeam.io/docs/profitability-benchmarks.md | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0 | I did not find a documented capability to push campaign or ad-set budget changes to advertising platforms through MCP/API or Northbeam. Apex sends attribution signals to Meta, while the documented MCP connection is read-only. | https://docs.northbeam.io/docs/apex-faqs.md https://docs.northbeam.io/docs/northbeam-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | I did not find evidence of direct bid-parameter changes through MCP/API or Northbeam. The Meta Custom Attribution guide requires automatic bidding and excludes non-autobid strategies, which does not establish a target-ROAS write capability. | https://docs.northbeam.io/docs/setting-up-a-meta-custom-attribution-campaign.md | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0 | I did not find an execution-linked pre/post analysis reporting revenue and spend effects for each campaign or ad-set bidding change. The reviewed Meta guide describes controlled Custom Attribution comparisons, without documenting that specific change-history workflow. | https://docs.northbeam.io/docs/setting-up-a-meta-custom-attribution-campaign.md | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.5 | Northbeam documents geographic holdouts and completed-test readouts containing incremental revenue, iROAS, and uncertainty. This is strong platform evidence, but I did not find a documented MCP/API operation retrieving geo-test results. | https://docs.northbeam.io/docs/announcements/incrementality.md https://docs.northbeam.io/interpreting-your-results.md | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0 | I did not find evidence of owned-media incrementality readouts for controlled email or leaflet tests through MCP/API or Northbeam. The reviewed incrementality guide describes advertising holdouts without identifying those owned-media test types. | https://docs.northbeam.io/docs/incrementality.md | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0 | I did not find evidence that Northbeam retrieves Meta Conversion Lift study results through MCP/API or its platform. The reviewed Meta integration documents Custom Attribution optimization tests, which do not establish Conversion Lift result ingestion. | https://docs.northbeam.io/docs/setting-up-a-meta-custom-attribution-campaign.md | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | Northbeam recommends testing high-spend channels with uncertain attribution, providing partial platform-workflow evidence for prioritization. I did not find a system-generated ranking of experiments using measured uncertainty or a corresponding MCP/API tool. | https://docs.northbeam.io/docs/use-cases/validate-with-incrementality.md | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.25 | Northbeam advertises automated self-service test design, region balancing, and validation, and its technical guide discusses power and spend requirements. This is partial platform evidence because I did not find a complete output specification covering treatment, control, power, duration, and required spend or an MCP/API design endpoint. | https://www.northbeam.io/blog/introducing-incrementality-by-northbeam-automated-lift-testing-you-can-trust https://docs.northbeam.io/docs/incrementality.md | Marketing collateral Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Northbeam publishes separate connection guides for ChatGPT and Claude, including Claude Code. Both document the server URL, authentication, prerequisites, and a verification prompt. | https://docs.northbeam.io/docs/connect-northbeam-to-chatgpt https://docs.northbeam.io/docs/connect-northbeam-to-claude | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | I did not find evidence of server-provided chart or table artifacts rendered natively by a compatible MCP client. The MCP documentation describes data retrieval and assistant summaries without documenting native visual outputs. | https://docs.northbeam.io/docs/northbeam-mcp-server-faqs | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | Northbeam publishes API request schemas and discovery endpoints for metrics, attribution models, and breakdowns, alongside metric definitions. This satisfies the API-equivalence rule, although I did not find a separate public MCP tool-schema catalog. | https://docs.northbeam.io/reference/post_data-export-1 https://docs.northbeam.io/docs/northbeam-api-data-export-1 https://docs.northbeam.io/docs/northbeam-metrics-101 | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The Data Exports API returns structured job status and downloadable CSV results delivered to customer-controlled storage. This satisfies the API-equivalence rule for reusable data, while Northbeam explicitly distinguishes MCP queries from bulk exports. | https://docs.northbeam.io/docs/northbeam-api-data-exports-overview https://docs.northbeam.io/docs/exporting-data | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 1 | Northbeam documents API-based automated pipelines and scheduled warehouse exports, as well as an authenticated Claude Code connection. The API-equivalence rule supports full credit for repeatable external automation, although scheduling the analysis remains the external runtime’s responsibility. | https://docs.northbeam.io/docs/use-cases/pull-data-into-your-warehouse.md https://docs.northbeam.io/docs/connect-northbeam-to-claude | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | Northbeam has a documented MMM based on time-series regression that supports channel planning. This is partial analytical-backbone evidence because I did not find confirmation of Bayesian estimation or that MCP recommendations execute that model. | https://docs.northbeam.io/docs/mixed-media-modeling.md | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | The technical MMM guide states that incrementality results calibrate its estimates. The incrementality announcement explicitly includes calibration for MMM, providing strong platform evidence under the categories 6–8 scoring rule. | https://docs.northbeam.io/docs/mixed-media-modeling.md https://docs.northbeam.io/docs/announcements/incrementality.md | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 0 | I did not find evidence of customer-visible MMM validation metrics such as R-squared, MAPE, posterior predictive checks, or holdout accuracy. The reviewed MMM guide discusses model-quality dependencies without documenting those diagnostic outputs. | https://docs.northbeam.io/docs/mixed-media-modeling.md | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | Northbeam describes a self-service MMM dashboard where users can adjust models and budget mixes. This is partial evidence because I did not find documentation of editable priors or an auditable calibration and configuration history. | https://www.northbeam.io/mmm-demo | Product page |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | I did not find evidence of controlled model-calibration changes with validation and version history through MCP or an equivalent API. The MCP guide explicitly limits access to read-only queries, and the reviewed public API documentation covers data operations rather than calibration. | https://docs.northbeam.io/docs/northbeam-mcp https://docs.northbeam.io/docs/northbeam-api-data-export-1 | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0 | Northbeam publishes numerous customer stories, including Timex, Dr Squatch, Ridge, and HexClad. I did not find evidence in the allowed sources establishing at least ten reference customers whose individual brand revenue exceeds one billion dollars. | https://www.northbeam.io/customer-stories | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Northbeam states that it is SOC 2 Type 2 compliant and makes the report available on request. This is strong public evidence of an independently audited security program, although the audit report itself was not reviewed. | https://www.northbeam.io/data-security | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | I did not find evidence that customers can choose between US and EU data residency. The GDPR statement discusses safeguards for transfers outside the EU or EEA without offering a selectable hosting geography. | https://www.northbeam.io/gdpr | Product page |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | Northbeam says most of its platform runs on Google Cloud Platform. I did not find evidence of customer-selectable deployment across AWS, GCP, and Azure, and export destinations do not establish that hosting choice. | https://www.northbeam.io/data-security https://docs.northbeam.io/docs/northbeam-api-data-exports-overview | Product page Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0.5 | Northbeam’s enterprise ChatGPT connection guide references enterprise credentials and an internal single-sign-on token, while its security page names Auth0 identity controls. This is partial evidence because I did not find a dedicated customer SSO setup guide confirming supported enterprise identity providers and protocols. | https://docs.northbeam.io/docs/connect-northbeam-to-chatgpt https://www.northbeam.io/data-security | Technical doc Product page |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | MCP access requires an authenticated Northbeam Pro or Enterprise account. I did not find a publicly accessible promptable MCP demo or ungated trial, and the published MMM+ demo requires a booked introduction. | https://docs.northbeam.io/docs/northbeam-mcp https://www.northbeam.io/mmm-demo | Technical doc Product page |
Recast
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 0.75 | The Insights API reports that the MCP can run include in-sample model fit (predicted vs. actual KPI) and a period summary, all filtered by date for the modeled KPI. Online sales are covered when they are the modeled KPI, but the docs do not describe a dedicated online sales report. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.75 | API examples use retail KPIs such as Walmart and Amazon revenue, sales data can come from the customer's Snowflake, BigQuery or Redshift, and the MCP reports the actual KPI for a date range. We did not find physical-store sales named explicitly as a supported KPI. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 0.75 | The compute_spend_summary report, which the MCP can run, returns spend per channel (e.g. Meta, search, paid social) at daily, weekly, monthly or total level. It reports spend only; we found no evidence of impressions or other media metrics, and the docs advise against modeling on them. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | The same spend summary covers every modeled channel, and Recast lists TV, radio, podcasts and direct mail among the offline channels it models. Only spend is reported, and we found no evidence covering out-of-home or print specifically. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The Data Guide lists revenue, profit, conversions and marketing qualified leads per day as supported KPIs, with new customer acquisition as a typical example. The API changelog confirms that reports run through the MCP return CPA and mCPA for conversion-based models. | https://docs.getrecast.com/docs/changelog | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | MCP reports take explicit start and end dates plus channel lists, and some take custom channel groupings and a daily, weekly or monthly granularity. Brands, markets and products are covered by picking a separate model (deployment), and we found no campaign filter. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The compute_insights_overview report, available through the MCP, returns each channel's incremental impact, ROI and marginal ROI for the period. These are MMM-based incremental estimates, not platform or last-click ROAS. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 1 | The same channel overview covers every modeled channel, and Recast says it measures TV, radio, podcasts and direct mail. The MCP therefore returns incremental ROI and impact for offline channels. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 1 | The MCP launch post features ranking last year's promotions by incremental value using the Spike Detail report, which estimates pull-forward and pull-back. A context variable summary report also shows how price affects marketing effectiveness. | https://getrecast.com/mcp-is-here/ | Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.25 | Recast estimates ROI for each channel on each day, but the model refreshes weekly (with on-demand refreshes) and the MCP reads those weekly deployments. We found no evidence of daily MMM model updates. | https://getrecast.com/recast-llm-information/ | Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | Reports available through the MCP include marketing vs. baseline, a baseline summary with spike effects, and a context variable summary. Together they split outcomes into baseline, media, promotions and non-media drivers. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 1 | The Optimizer API (POST /optimizations) runs Recast's optimizer to calculate channel budgets for a chosen objective, target and period. Under the evaluation rules, API features count the same as MCP features. | https://docs.getrecast.com/docs/the-optimizer-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 1 | The Forecaster API (POST /forecasts) takes a daily budget per channel and returns forecasted outcomes and ROI with uncertainty quantiles. Plans API forecasts also return expected outcome and ROI for a saved budget. | https://docs.getrecast.com/docs/the-forecaster-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 1 | The compute_spend_response_curves report returns impact, ROI/CPA and marginal ROI at different spend levels for selected channels. The MCP launch post shows looping spend response curves across channels to rank how much room each has before saturation. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 1 | The Plans API accepts a budget patch, such as raising Meta by 10%, as a new plan version and forecasts its outcome. The fixed-spend performance report also estimates a channel's performance at a given spend level. | https://docs.getrecast.com/docs/the-plans-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 1 | The Optimizer API form includes the target, committed (minimum) spend per channel, drop days, lower-funnel caps and dates, and any Optimizer UI setting can be passed through it. GET /optimizations/[id] returns the full form, so the constraints applied can be checked. | https://docs.getrecast.com/docs/the-optimizer-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 0.75 | Through the API you can read the auto-generated default plan (business as usual) and alternative plans or forecasts, each with comparable spend and outcome figures, plus counterfactuals of planned vs. actual spend. We did not find a single endpoint that returns a side-by-side scenario vs. baseline comparison. | https://docs.getrecast.com/docs/the-plans-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 1 | Optimizer objectives include an outcome target, an efficiency target (ROI/CPA) and maximizing profit, with a required confidence level. These are passed through the target and objective fields of the Optimizer API form. | https://docs.getrecast.com/docs/optimizer | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 1 | GET endpoints for optimizations, forecasts and plans return saved runs with their IDs, input forms, results and CSV downloads. Plan versions keep a history of budgets and forecasts. | https://docs.getrecast.com/docs/the-plans-api | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.25 | We found no evidence of campaign or ad set ROI through the MCP or API. The platform's sub-channel analysis gives campaign-level performance indices relative to the parent channel, which are not time-varying and not absolute incremental ROAS. | https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.25 | We found no evidence that the MCP returns last-click or ad platform ROAS for campaigns or ad sets. The launch post shows an LLM joining Recast's channel ROI with the user's own Google Ads or Meta export, which only partly covers this. | https://getrecast.com/mcp-is-here/ | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0 | We found no evidence of marginal ROI at campaign or ad set level through the MCP, the API or the wider platform. Marginal ROI is documented at channel level only. | https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | We found no evidence of campaign or ad set daily budget recommendations through the MCP or API. The docs describe Optimizer downloads splitting a parent channel's recommended spend across sub-channels mostly by historical share, which only partly covers this. | https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0 | We found no evidence of bid value or target ROAS recommendations through the MCP, the API or the wider platform. The documented optimizer outputs are budgets. | https://docs.getrecast.com/docs/optimizer | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0 | We found no evidence that the MCP or platform can push budget changes to Meta, Google or other ad platform APIs. The documented MCP actions only cover deployments and reports. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | We found no evidence that the MCP or platform can push bidding changes such as target ROAS to ad platforms. The documented MCP actions only cover deployments and reports. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | We found no evidence of pre/post analysis of executed bidding changes at campaign or ad set level. The Time Period Comparison report compares ROI, mROI and effect across two periods at channel level, and the docs suggest it for before/after checks of optimizations. | https://docs.getrecast.com/docs/reporter | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | The compute_experiments_summaries report ('experiment results summary') is available through the Insights API, which the MCP uses, and GeoLift test results are loaded into the MMM as experiments. GeoLift itself has no MCP or API access, and the summary's fields are not documented. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.75 | Recast's experiments doc lists direct mail matchback with holdout testing among the test types that can be loaded, and experiment results can be retrieved through the experiments summary report. We found no evidence covering leaflet or email tests specifically. | https://docs.getrecast.com/docs/experiments | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.75 | The experiments doc names in-platform lift studies such as Meta Conversion Lift as valid inputs, and experiment results can be retrieved through the experiments summary report. What that report returns is not documented in detail. | https://docs.getrecast.com/docs/experiments | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.75 | The Reporter docs suggest using the Daily Spend vs KPI report to find highly correlated channels 'in order to prioritize testing', and Channel Uncertainty Drivers shows why a channel's estimate is uncertain; both are in the API but excluded from the MCP. They support prioritization, but we found no automated recommendation of specific tests. | https://docs.getrecast.com/docs/reporter | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | GeoLift by Recast designs geo tests with market selection, power analysis, test length, budget and confidence interval previews. The docs state the MCP cannot act on GeoLift, and we found no GeoLift API. | https://geolift-docs.getrecast.com/docs/how-to-design-an-optimal-recast-geolift-test | Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | The MCP Setup Guide gives step-by-step OAuth connection instructions for ChatGPT (web and desktop), Codex, Claude (web, desktop and Claude Code), Gemini CLI and Snowflake Cortex. It also covers the plugin bundling the MCP and API skills. | https://docs.getrecast.com/docs/mcp-setup-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | We found no evidence of server-provided tables or chart artifacts. The MCP docs state it cannot return graphs and suggest asking the AI client to draw charts from the data. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | The MCP docs list the tools (list_deployments, get_deployment, search_tools, run_report, get_report, download_report), and search_tools lets the agent find reports by keyword. Each report and metric is documented in the Reporter docs and Reporting Standards glossary. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The MCP's download_report tool returns report data as CSV, and the Optimizer, Forecaster, Plans and Insights APIs return JSON plus CSV downloads. Outputs can also be exported to the client's S3 bucket or data warehouse. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 1 | The docs recommend the API and plugin skills for scheduled jobs such as pulling data every Tuesday after the model refreshes, and document MCP setup in Codex CLI, Claude Code and Snowflake Cortex agents. The launch post also describes automating weekly pulls into agents and scripts. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 1 | The MCP reports come from Recast's Bayesian hierarchical time series MMM, fitted with Hamiltonian Monte Carlo in Stan and using time-varying ROIs. The docs describe the model structure in detail. | https://docs.getrecast.com/docs/recast-model-technical-documentation | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Lift and geo test results enter the MMM as priors on cumulative ROI over the test dates, with options for the distribution type. Multichannel impact tests can also calibrate channels tested together. | https://docs.getrecast.com/docs/how-recast-applies-incrementality-tests-to-the-mmm | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | The platform's Backtests tab reports out-of-sample forecast error (CRPS) at several horizons, and backtest summaries and in-sample fit are available as API reports. We did not find R2 or MAPE named explicitly. | https://docs.getrecast.com/docs/backtests | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | The Configuration page lets customers view priors and model settings in the UI. Changes are requested through a 'Request a configuration change' button that emails Recast support, so this is not self-serve editing. | https://docs.getrecast.com/docs/configure | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We found no evidence that agents can propose or apply calibration changes through the MCP or API. Configuration and experiment reports are read-only, and changes go through Recast staff. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | The site names about 18 customers, including Block (Square, Cash App, Afterpay), Canva and Hasbro, which are clearly $1B+ revenue brands. Most other named references (e.g. PODS, Harry's, LA Times, KOHO) are smaller or uncertain, so we found fewer than 10 at $1B+. | https://getrecast.com/recast-case-studies/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Recast states that it is SOC 2 compliant and shows an AICPA SOC badge on its product pages. The report type (Type I or II) is not stated, and we found no evidence of ISO 27001. | https://getrecast.com/mmm/ | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | We found no evidence of a choice between US and EU data residency on the website or in the docs. Documented data exchange uses Recast's own AWS S3 buckets without a stated region option. | https://docs.getrecast.com/docs/exchanging-data-with-recast | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We found no evidence of a choice between AWS, GCP and Azure for hosting. Recast describes itself as a cloud-native SaaS on AWS, although it can take data from BigQuery, GCS and Snowflake. | https://getrecast.com/recast-llm-information/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | We found no evidence of enterprise SSO (SAML, Okta, Microsoft Entra) for the Recast app. The MCP uses OAuth tied to a Recast login and the API uses personal access tokens. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We found no evidence of a public, promptable MCP demo or sandbox; the MCP docs say a Recast login is needed to authenticate. The MMM site offers 'Book a Demo', and only the separate GeoLift product has a free tier. | https://docs.getrecast.com/docs/the-recast-mcp | Technical doc |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | Reporter exposes observed KPI totals for selected dates, and Recast documents separate DTC and Amazon revenue models. The MCP provides report access for configured online-sales KPIs. | https://docs.getrecast.com/docs/reporter https://getrecast.com/lume-case-study/ https://docs.getrecast.com/docs/the-recast-mcp | Technical doc Marketing collateral Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 1 | Recast documents retail-sales models and ingestion from customer warehouses, including BigQuery and Snowflake. Its API exports observed model-input KPIs, enabling physical-store sales reporting when that KPI is configured. | https://getrecast.com/lume-case-study/ https://docs.getrecast.com/docs/data-guide https://docs.getrecast.com/docs/roi-models-data-export https://docs.getrecast.com/docs/the-insights-api | Marketing collateral Technical doc Technical doc Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 0.75 | The API provides spend reports for configured digital channels. I did not find equivalent reporting of delivery metrics such as clicks, impressions and reach across the major advertising platforms. | https://docs.getrecast.com/docs/the-insights-api https://docs.getrecast.com/docs/data-guide https://docs.getrecast.com/docs/does-recast-use-last-click-metrics | Technical doc Technical doc Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | Spend reporting through the API applies to configured offline channels, with Recast explicitly describing TV and radio coverage. I did not find documentation establishing the full TV, out-of-home, radio and print delivery-metric reporting requirement. | https://docs.getrecast.com/docs/the-insights-api https://getrecast.com/recast-llm-information/ | Technical doc Marketing collateral |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The data specification supports profit, conversions, leads and customer-acquisition KPIs. Observed KPI reporting and model-input downloads expose these configured outcomes through the API. | https://docs.getrecast.com/docs/data-guide https://docs.getrecast.com/docs/reporter https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | Report requests support date ranges, deployment selection and channel groupings with multiple time granularities. I did not find a complete API filter schema covering brand, market, product and campaign dimensions together. | https://docs.getrecast.com/docs/the-insights-api https://getrecast.com/recast-llm-information/ | Technical doc Marketing collateral |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | Recast defines channel ROI as incremental return and exports attributable revenue. These measures are available through the Insights API for configured digital channels. | https://docs.getrecast.com/docs/roi-models-data-export https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 1 | Recast measures incremental performance across online and offline media, including TV and radio. The same channel reports and API access cover offline-channel ROI and attributable revenue. | https://getrecast.com/recast-llm-information/ https://docs.getrecast.com/docs/roi-models-data-export https://docs.getrecast.com/docs/the-insights-api | Marketing collateral Technical doc Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 1 | Recast documents promotion effects separately from paid-media effects and supports configured price effects. Spike and context reports are accessible through MCP reporting. | https://docs.getrecast.com/docs/reporter https://docs.getrecast.com/docs/the-recast-mcp | Technical doc Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0 | Recast documents weekly model re-estimation and a configurable refresh schedule. I did not find evidence of daily MMM measurement refreshes, and daily output granularity alone does not satisfy this criterion. | https://getrecast.com/mmm/ https://docs.getrecast.com/docs/the-refresh-tracker | Product page Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | Reporter decomposes observed outcomes into paid media, baseline and promotional spikes, with additional configured context effects. The API exposes the corresponding reports. | https://docs.getrecast.com/docs/reporter https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 1 | POST /optimizations invokes Recast’s optimization engine and returns downloadable results. Its form carries objectives, model choices and planning constraints, satisfying the API-equivalence rule. | https://docs.getrecast.com/docs/the-optimizer-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 1 | The Forecaster API accepts daily channel budgets and returns modeled outcomes. Forecasts must start immediately after the model’s last observed day, a documented limit on supported plans. | https://docs.getrecast.com/docs/the-forecaster-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 1 | The spend-response report returns impact and returns across spend levels, while channel reporting includes marginal returns. Recast exposes these reports through its MCP reporting tools. | https://docs.getrecast.com/docs/reporter https://docs.getrecast.com/docs/roi-models-data-export https://docs.getrecast.com/docs/the-recast-mcp | Technical doc Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 1 | Users can change channel-level daily budget inputs and submit another forecast through the API. The engine calculates the resulting modeled outcomes with the selected deployments and assumptions. | https://docs.getrecast.com/docs/the-forecaster-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 1 | The Optimizer supports dated channel constraints, minimum and maximum allocations, fixed totals and committed spend. Its API exposes the same settings through the submitted and retrieved form. | https://docs.getrecast.com/docs/optimizer https://docs.getrecast.com/docs/the-optimizer-api | Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 1 | The Plans API retrieves versioned budgets and forecasts, including planned-budget and actual-budget counterfactuals. An agent can select an explicit baseline and alternative versions and compare their returned spend and outcome estimates. | https://docs.getrecast.com/docs/the-plans-api | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 1 | Optimizer objectives include outcome targets, efficiency targets and profit maximization using margin or acquisition value. The API accepts these objective and target settings and returns an optimized budget. | https://docs.getrecast.com/docs/optimizer https://docs.getrecast.com/docs/the-optimizer-api | Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 1 | The Plans API lists saved plans and versions with stable identifiers. Version configuration, daily budgets and associated forecasts provide the saved assumptions, inputs and outputs. | https://docs.getrecast.com/docs/the-plans-api | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.25 | Recast’s sub-channel feature estimates campaign efficiency relative to its parent channel. I did not find evidence of absolute incremental revenue and ROAS for individual campaigns and ad sets through MCP or API. | https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0 | Recast illustrates combining model estimates with external Meta or Google exports. I did not find a documented campaign-and-ad-set comparison returning incremental, last-click and ad-platform ROAS together. | https://getrecast.com/mcp-is-here/ https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Marketing collateral Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0 | Recast documents marginal returns at channel level and relative efficiency for sub-channels. I did not find documented marginal incremental ROAS for individual campaigns and ad sets in the platform or its interfaces. | https://docs.getrecast.com/docs/roi-models-data-export https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.75 | Optimizer downloads include sub-channel budget allocations, and the API exposes optimization downloads. This is partial evidence because daily recommendations for both campaigns and ad sets are not explicitly documented. | https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast https://docs.getrecast.com/docs/the-optimizer-api | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0 | The documented optimization outputs allocate spend across channels and sub-channels. I did not find evidence of recommended campaign or ad-set bidding parameters such as target ROAS. | https://docs.getrecast.com/docs/optimizer https://docs.getrecast.com/docs/how-to-use-sub-channels-in-recast | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0 | The documented API workflow returns optimized budgets and CSV downloads. I did not find an endpoint or platform feature that writes daily campaign or ad-set budgets directly to Meta or Google Ads. | https://docs.getrecast.com/docs/the-optimizer-api | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | Recast documents reporting, planning and optimization interfaces. I did not find an MCP tool, API endpoint or broader-platform feature that executes bid changes in advertising platforms. | https://docs.getrecast.com/docs/the-recast-mcp https://docs.getrecast.com/docs/the-optimizer-api | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0 | Recast offers channel-level comparisons between time periods. I did not find evidence linking executed campaign or ad-set bidding changes to an actual revenue-and-spend pre/post analysis. | https://docs.getrecast.com/docs/reporter | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 1 | Recast records ingested geographic lift tests with channel, dates, outcome estimate and uncertainty. Its API exposes experiment summaries, supporting retrieval of completed tests incorporated into the MMM rather than direct GeoLift execution. | https://docs.getrecast.com/docs/experiments https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 1 | Supported imported experiments include direct-mail holdout tests, with results displayed in the model’s experiment list. The experiment-summary API provides access to these ingested results, conditional on the test having been supplied to Recast. | https://docs.getrecast.com/docs/experiments https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 1 | Recast explicitly supports incorporating Meta Conversion Lift results and displaying their estimates and uncertainty. The API’s experiment-summary report exposes ingested experiment results, without establishing a direct Meta import connection. | https://docs.getrecast.com/docs/experiments https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | Recast describes building an experimentation roadmap around uncertain or strategically important channels with its team. I did not find a documented MCP or API tool that recommends specific experiments, so the evidence supports only a partial broader-platform workflow. | https://getrecast.com/the-recast-process/ | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | GeoLift’s design tool selects treatment markets and synthetic controls and evaluates duration, required spend and statistical power. I did not find MCP or API access to this design engine, so this receives the broader-platform score. | https://geolift-docs.getrecast.com/docs/how-to-design-an-optimal-recast-geolift-test https://docs.getrecast.com/docs/the-recast-mcp | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | The setup guide provides separate connection instructions for ChatGPT and Claude. It specifies the hosted MCP address, OAuth authorization and Recast sign-in. | https://docs.getrecast.com/docs/mcp-setup-guide | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | Recast states that its MCP does not return graphs, and its API report downloads omit plots. I did not find a documented server-provided visual table or chart artifact rendered natively in an MCP client. | https://docs.getrecast.com/docs/the-recast-mcp https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | Recast documents named MCP tools and keyword-based report discovery. Its Reporter documentation explains the meaning and interpretation of returned metrics. | https://docs.getrecast.com/docs/the-recast-mcp https://docs.getrecast.com/docs/reporter | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | Reports expose downloadable CSV data, with API responses identifying available download keys. This supports spreadsheets and downstream tools, although plot-only content is excluded from downloads. | https://docs.getrecast.com/docs/the-insights-api | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 1 | Recast documents external-agent and scheduled-script workflows using its API skills. Published examples show authenticated submission, polling and CSV retrieval for repeatable analyses. | https://docs.getrecast.com/docs/the-recast-mcp https://docs.getrecast.com/docs/the-forecaster-api | Technical doc Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 1 | Recast documents a Bayesian MMM, and its interfaces retrieve results from identified model deployments. This supports model-backed answers rather than LLM-invented estimates, while the underlying statistical calculations remain probabilistic. | https://docs.getrecast.com/docs/recast-model-technical-documentation https://docs.getrecast.com/docs/the-recast-mcp | Technical doc Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Recast incorporates lift-test estimates and their uncertainty as time-specific Bayesian constraints. Documentation covers geographic tests, direct-mail holdouts and advertising-platform lift studies. | https://docs.getrecast.com/docs/experiments | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | Backtests display out-of-sample accuracy and CRPS across forecast horizons. The Insights API includes backtest and in-sample model-fit reports for programmatic validation reporting. | https://docs.getrecast.com/docs/backtests https://docs.getrecast.com/docs/the-insights-api | Technical doc Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0.5 | Customers can inspect priors in the Configuration UI but request changes through support. Operators can edit an external prior-setting spreadsheet and launch models, which is partial evidence rather than full in-app self-service configuration. | https://docs.getrecast.com/docs/configure https://operators.getrecast.com/docs/building-your-first-model-demo | Technical doc Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0.5 | The operator API launches models from a prior-settings sheet and supports holdout, stability and parameter-recovery validation runs with retrievable run records. I did not find a complete documented calibration-change approval and immutable version-history workflow, so support is partial. | https://operators.getrecast.com/docs/run-launching-api-endpoint | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Recast publishes named references including Block, Canva and PODS and describes serving Fortune 500 advertisers. I did not find company-site evidence verifying at least ten distinct reference customers each exceeding $1 billion in annual revenue. | https://getrecast.com/recast-llm-information/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Recast explicitly states SOC 2 compliance and displays an AICPA SOC badge on its product website. This supports the stated security criterion, although I did not inspect the underlying audit report or its coverage dates. | https://getrecast.com/ | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | I did not find public documentation of a customer-selectable US-versus-EU data-residency option. The reviewed architecture description identifies AWS hosting without establishing a regional choice. | https://getrecast.com/recast-llm-information/ | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | Recast identifies AWS as its SaaS hosting environment and supports several external data sources. I did not find evidence that customers can choose to deploy the platform on AWS, GCP or Azure. | https://getrecast.com/recast-llm-information/ https://docs.getrecast.com/docs/data-guide | Marketing collateral Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | Recast’s implementation documentation explicitly supports enterprise SAML SSO. It provides identity-provider configuration details and the process for completing the integration. | https://docs.getrecast.com/docs/onboarding-faq | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | I did not find a publicly accessible, promptable MCP or API trial without sales gating. GeoLift has a self-service free trial, but the MCP requires a provisioned Recast account and excludes GeoLift actions. | https://getrecast.com/geolift-by-recast/ https://docs.getrecast.com/docs/the-recast-mcp | Product page Technical doc |
Sellforte
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The MCP reporting guide lists online sales as a reportable area, with totals by week, month, quarter or year for any period and comparison periods. It includes a worked example of monthly online sales for 12 months with a quarter-on-quarter comparison. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 1 | The MCP reporting guide says physical store sales work exactly like online sales and shows an example splitting sales into online and store by quarter. Separate support docs describe Sellforte fetching data from the customer's storage or data warehouse. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The MCP reports spend, impressions, clicks, CPM, CPC, conversions and conversion value for connected platforms including Meta, Google, TikTok, Amazon and Criteo. Campaign and ad set names follow the ingested ad platform structure. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 1 | TV, out of home, radio, print, cinema, direct mail and leaflets sit in the same media dataset as digital and are reported with the same prompts. Spend is always available, while impressions, contacts or GRPs depend on what the agency or media owner delivers. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The MCP reports every KPI configured in the customer's environment, such as contribution margin, gross profit, orders, new customers and conversions. The guide includes an example returning revenue, contribution margin and new customers by quarter. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 1 | The MCP resolves user wording to exact values in the data model and filters by date, market, brand, product, sales channel, customer type, platform and campaign. Granularity can be total, day, week, month, quarter or year, with relative or exact comparison periods. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The MCP returns MMM incremental sales and incremental ROAS for every digital channel, platform, campaign and ad set in the model. The guide states these come from the MMM rather than from last click, with GA4 and platform ROAS shown only for comparison. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 1 | The reporting guide states offline media is measured in the same model as digital, so TV, OOH, radio, print, cinema and direct mail get the same incremental revenue and incremental ROAS. Example prompts rank digital and offline channels together by incremental ROI. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 1 | The MCP reports promo investment, incremental sales and margin from promotions, and ROI and margin ROI by promotion type. The guide says promotions are separated inside the model so media ROIs exclude promotion-driven revenue. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 1 | The MCP reporting guide says Sellforte typically models at daily granularity and results update with the daily data pipeline, so incremental ROAS can be requested by day. The MCP also returns the actual refresh timestamp; the Media Optimizer FAQ separately states MMM results update every day. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | The MCP decomposes each KPI into base, media (per channel and campaign), promotions and other modeled drivers such as weather. The guide states the components sum to the total KPI and gives an example explaining the biggest monthly change. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 1 | The channel-level MCP guide says an optimized plan request runs Sellforte's optimization engine, reallocating a given budget across channels for a given period using the MMM response curves. The result is saved as a real Media Optimizer scenario. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 1 | The Reference scenario template takes a booked plan exactly as it is, week by week and channel by channel, and forecasts it. The guide lists expected sales, incremental sales and ROI as the outputs for a user's own plan. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.5 | We found no evidence that the MCP returns marginal ROI or response curve data; the MCP guides only say plans are built on the model's response curves. The broader platform documents miROAS at channel, campaign and ad set level in Sellforte Performance, and the Media Optimizer uses diminishing returns curves. | https://sellforte.com/support/working-with-miroas-in-performance | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 1 | The MCP simulates budget changes either as an even scale or cut across channels or as an optimized plan at the new budget. Example prompts include simulating cuts of 5, 10 and 20 percent and the effect of a 500 000 increase. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 1 | The MCP passes planning period, total budget, channel minimums and maximums, locked channels and channel-level changes to the optimizer. The guide states constraints come back resolved in the result so users can check what the optimizer was allowed to do. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 1 | Naming one scenario as the baseline returns the others against it with absolute and percentage differences per channel and in total for media investment, incremental sales and ROI. With three or more scenarios the values are shown side by side. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 1 | The MCP guide lists four target modes: total sales target, incremental sales target, target ROI and target CPA, where the optimizer works out the budget and mix to reach the goal. Target CPA appears only in the MCP guide, not in the Media Optimizer UI article we read. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 1 | Saved scenarios are retrievable through the MCP with identifier, planning window, resolved total budget, template, optimization mode, reference period and media performance assumption. Plans can be filtered by period and are flagged as stale against the last model refresh. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 1 | The MCP reporting guide states incremental sales and incremental ROAS are available for every digital channel, platform, campaign and ad set in the model. Results can also be filtered by campaign name and objective. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 1 | The MCP reporting guide lists incremental ROAS, GA4 attributed (last click) sales and ROI, and ad platform revenue and ROAS side by side for channels, campaigns and ad sets. Example prompts compare incremental ROI with Meta-reported ROAS for paid social campaigns. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.5 | We found no evidence in the public MCP guides of campaign or ad set miROAS through the MCP. The platform's Performance module documents miROAS as the return on the next unit of spend at campaign, ad set and channel level. | https://sellforte.com/support/working-with-miroas-in-performance | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | We found no evidence of campaign or ad set budget recommendations through the MCP; the public MCP planning guide covers channel, market and week level only. Sellforte Performance recommendations give daily budget changes per Google campaign and Meta ad set. | https://sellforte.com/support/performance-recommendations-overview | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.5 | We found no evidence of bid recommendations through the MCP. Sellforte Performance recommends target ROAS and other bidding parameters per campaign or ad set, and a 2026-08-19 update added target CPA recommendations for Google. | https://sellforte.com/support/performance-recommendations-overview | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | We found no evidence that the MCP pushes budget changes; the MCP FAQ says it does not change campaigns automatically and points to Sellforte Activate. The platform documents applying budget recommendations to Google, Meta and TikTok via their APIs after user confirmation. | https://sellforte.com/support/how-to-apply-a-performance-recommendation | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.5 | We found no evidence that the MCP pushes bidding changes to ad platforms. The platform documents applying target ROAS and bid recommendations through the ad platform APIs from Sellforte Performance after user confirmation. | https://sellforte.com/support/how-to-apply-a-performance-recommendation | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.5 | We found no evidence of pre/post analysis of executed changes through the MCP; the MCP page presents Change Intelligence as a separate product step. Change Intelligence detects target ROAS, bid and budget changes and compares pre and post periods against a counterfactual at campaign and ad set level. | https://sellforte.com/support/change-intelligence-overview | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | The MCP connection guide says the MCP lets users query their marketing mix and incrementality data, and the MCP page lists experiment data among its capabilities, but no MCP guide documents geo test retrieval. The platform's Geo Lift module stores results with incremental sales, iROAS and credible intervals. | https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.75 | Only the general MCP claim of access to experiment data was found, with no MCP guide covering A/B tests. The platform's A/B Test type analyzes uploaded test and control series for any grouping, such as customer segments or store tiers, though email and leaflet tests are not named. | https://sellforte.com/support/sellforte-experiments-a/b-test | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.75 | Only the general MCP claim of access to experiment data was found, with no MCP guide covering lift studies. The platform imports Meta Conversion Lift exports, derives an iROAS distribution with a 90% credible interval and stores the result in the Experiment Library. | https://sellforte.com/support/sellforte-experiments-conversion-lift | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | We found no evidence of experiment prioritization through the MCP. Platform docs only advise prioritizing channels where a test could change spend decisions and flagging surprising model results for testing, without a ranking feature based on uncertainty. | https://sellforte.com/support/how-to-conduct-a-geo-lift-experiment | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.25 | We found no evidence of test design through the MCP. The Geo Lift design guide covers objectives, regions, treatment type (holdout, -50% or +100%), balanced test and control split and typical 4 to 8 week duration, but we found no power or MDE calculation. | https://sellforte.com/support/how-to-conduct-a-geo-lift-experiment | Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 0.5 | The connection guide gives step-by-step instructions for Claude (Settings, Connectors, Add custom connector, EU or US URL, WorkOS sign-in and organization selection). For ChatGPT it only says setup follows a similar pattern under Connectors or Plugins, without specific steps. | https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 1 | Both MCP guides state that Sellforte renders its own tables and charts inside supported clients, so trends and comparisons appear as visuals. The MCP page lists rendering tables and visualizations among the connector's capabilities. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | The MCP page says connected clients discover the Sellforte tools automatically, and the reporting guide says the MCP reads datasets, metrics, dimensions and date range and returns the exact schema on request. The guide defines media and incremental metrics, although we found no public per-tool reference. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | Both MCP guides state that the underlying rows come back as structured data that the client can reuse for spreadsheets, slides or further calculations. We found no MCP file download; CSV and PNG exports are documented for the in-product Sellforte AI chat. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | The MCP page describes building MMM agents that call the optimizer when budgets change and says the MCP works with any MCP client, calling it an early step. We found no documentation of scheduled or headless invocation, and the documented authentication is interactive per-user OAuth. | https://sellforte.com/mcp | Product page |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 1 | Sellforte states its MMMs are built with a Bayesian approach using priors, and the MCP answers come from that model. The causal attribution page describes Bayesian modeling combining MMM data, experiments and attribution. | https://sellforte.com/marketing-mix-modeling | Product page |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Sellforte states its MMMs are calibrated with incrementality tests (conversion lift studies, geo lift tests and shutdown tests) as well as attribution data, using Bayesian priors. Its support tutorial explains that informative ROI priors are built from sources such as incrementality tests and geo experiments. | https://sellforte.com/marketing-mix-modeling | Product page |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | The Model Validation view reports R2, MAPE, bias and max APE by segment and granularity, with model version selection, in-sample vs actual charts and stated thresholds. It also describes held-out accuracy, backtesting and experiment comparison; posterior predictive checks and MCP access to these metrics were not found. | https://sellforte.com/support/understanding-model-validation | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 0 | We found no evidence that customers can inspect or edit priors or other calibration settings in a self-serve UI. Channel mapping rules are documented publicly and a model version selector exists, but the docs imply Sellforte manages model configuration. | https://sellforte.com/support/advertising-channel-mapping-in-sellforte | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We found no evidence that agents can propose or apply calibration changes through the MCP. The MCP FAQ describes current uses as reading measurement data and working with optimizer scenarios. | https://sellforte.com/mcp | Product page |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 1 | Sellforte names customers such as C&A, Lidl, Tchibo, Douglas, KiK, Fressnapf, Intersport, Bonprix, Azzas 2154, PPG, Peek & Cloppenburg, Gigantti (Elkjop), Paysafe and Fazer across its homepage, case studies and customer news. Around 14 appear to be $1B+ revenue brands, although several are logo or quote only and revenue sizes were not verified against annual reports. | https://sellforte.com/ | Product page |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Sellforte states its platform, software and processes are regularly audited by NIXU (part of DNV), a third-party cybersecurity auditor. We found no evidence of SOC 2 or ISO 27001 certification, and no audit report or scope is published. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 1 | The security page says customer data is in the EU unless agreed otherwise, with a choice of Dublin or Frankfurt. The MCP connection guide provides separate Europe and United States server URLs matching the customer's environment, though a US hosting region is not named on the security page. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 1 | The security FAQ says Sellforte primarily uses AWS, with Microsoft Azure and Google Cloud possible at separate pricing. The pricing page lists multi-cloud AWS or GCP, and we found no Azure hosting subprocessor in the DPA annex. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | The security page states customers can set up authentication with their own SSO system and that Azure AD (Entra ID) SSO and Google SSO are supported. The pricing page lists SSO on the Full-Funnel MMM and Enterprise plans. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We found no evidence of a public, promptable MCP demo; MCP access requires a subscription with the MMM module and a per-user invitation or waitlist place. A public demo environment with an in-product AI chat is available without sales contact, but it does not use the MCP. | https://sellforte.com/mcp | Product page |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The reporting guide documents period-based online sales queries through MCP. Monthly totals and period comparisons are explicitly supported. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 1 | MCP reports physical-store sales supplied from the customer’s data warehouse. The guide confirms the same periods and filters as online sales. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | MCP returns paid-media spend, impressions, clicks and conversion metrics. Meta, Google and TikTok are explicitly covered. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 1 | MCP reports offline spend and media metrics across TV, out-of-home, radio and print. Impression or contact availability depends on supplied data. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | MCP exposes configured business KPIs beyond revenue. The guide names contribution margin, orders and new customers. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 1 | MCP supports date, brand, market, product and campaign filtering and grouping. Results can use explicitly requested time granularity. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The guide documents MMM-based incremental revenue and ROI for digital channels. These are distinguished from platform and last-click attribution. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 1 | Offline channels receive incremental revenue and ROI through MCP. Documented examples include TV, out-of-home, radio and print. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 1 | MCP reports promotional incremental sales and margin. Promotion and pricing contributions are separated from media-driven results. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 1 | The reporting guide states that modeled results update with the daily pipeline. MCP exposes daily incremental ROI and the environment’s actual refresh timestamp. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | MCP decomposes outcomes into baseline, media, promotions and other modeled drivers. The guide states that these contributions reconcile to the total KPI. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 1 | MCP invokes the optimizer to allocate a stated budget across channels and a planning period. The documented optimized template maximizes the modeled outcome. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 1 | The reference-scenario template forecasts a supplied weekly channel plan without reallocating it. MCP returns expected sales, incremental sales and ROI. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.5 | Sellforte documents channel marginal returns and response curves in its platform. I found no explicit MCP or API retrieval of both marginal-return values and spend-response data. | https://sellforte.com/support/working-with-miroas-in-performance https://sellforte.com/support/media-optimizer-overview https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc Technical doc Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 1 | MCP simulates budget increases and cuts. It supports proportional changes or reoptimization at the revised budget. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 1 | MCP passes budget limits, channel restrictions and planning weeks to the optimizer. Returned results expose the resolved constraints. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 1 | MCP compares alternative scenarios with a named baseline. The comparison includes spend, incremental sales, ROI and channel-level differences. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.75 | The MCP guide gives a revenue-target planning prompt and supports retrieving target-based scenarios. However, it also says target modes are set in the UI, leaving direct MCP target-plan creation ambiguous. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 1 | MCP retrieves saved scenario identifiers, inputs and results. Planning assumptions and model-refresh freshness flags are documented. | https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 1 | The reporting guide explicitly covers incremental sales and ROI for campaigns and ad sets. Access is conditional on that granularity being present in the customer’s model. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 1 | The MCP guide lists incremental, GA4 last-click and ad-platform returns for campaigns and ad sets. These can be retrieved together for comparison. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.5 | Performance exposes modeled miROAS for individual campaigns and ad sets. I found no explicit documentation that this marginal metric is retrievable through MCP or a customer-callable API. | https://sellforte.com/support/working-with-miroas-in-performance https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Technical doc Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | Performance Recommendations provides daily campaign and ad-set budget recommendations. I found no documented MCP or customer-callable API operation returning those recommendations. | https://sellforte.com/support/performance-recommendations-overview https://sellforte.com/mcp | Technical doc Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.5 | Performance Recommendations documents optimal bidding targets, including Google target ROAS. I found no explicit MCP or customer-callable API access to these bidding recommendations. | https://sellforte.com/support/performance-recommendations-overview https://sellforte.com/mcp | Technical doc Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | The platform can apply recommended budget changes directly to connected ad platforms. I found UI execution evidence but no exposed Sellforte MCP or customer-callable API action for triggering the change. | https://sellforte.com/support/how-to-apply-a-performance-recommendation https://sellforte.com/mcp | Technical doc Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.5 | Sellforte documents applying bidding changes to Google campaigns and Meta ad sets. I found no MCP or customer-callable Sellforte API operation for initiating those writes. | https://sellforte.com/support/how-to-apply-a-performance-recommendation https://sellforte.com/mcp | Technical doc Product page |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.5 | Change Intelligence measures spend and incremental sales before and after bid or budget changes, with a counterfactual. I found no documented MCP or API retrieval of this change-specific analysis. | https://sellforte.com/support/change-intelligence-overview https://sellforte.com/mcp | Technical doc Product page |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 0.75 | The MCP product page states that experiment data is accessible, and Geo Lift documentation establishes intervention-linked results. I found no Geo Lift-specific MCP example or response definition, so the interface evidence is partial. | https://sellforte.com/support/sellforte-experiments-geo-lift https://sellforte.com/mcp | Technical doc Product page |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0.75 | Sellforte supports controlled A/B test results, while its MCP product page promises experiment-data access. I found no owned-media-specific MCP retrieval example, so coverage of that test type remains partially evidenced. | https://sellforte.com/support/sellforte-experiments-a/b-test https://sellforte.com/mcp | Technical doc Product page |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.75 | Sellforte stores and reports Meta Conversion Lift results and advertises experiment-data access through MCP. I found no conversion-lift-specific MCP retrieval example or schema, so this receives partial interface credit. | https://sellforte.com/support/sellforte-experiments-conversion-lift https://sellforte.com/mcp | Technical doc Product page |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0 | I found no documented software workflow that uses measurement uncertainty to prioritize specific experiments through MCP, API or the broader platform. The reviewed materials describe experiment analysis and manual design guidance, which do not establish this capability. | https://sellforte.com/experiments-agent https://sellforte.com/support/how-to-conduct-a-geo-lift-experiment | Product page Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.25 | The Experiments Agent page claims automated experiment design, providing partial platform evidence. I found no MCP or API workflow, or platform demonstration producing treatment, control, power or minimum detectable effect, duration and required spend together. | https://sellforte.com/experiments-agent https://sellforte.com/support/how-to-conduct-a-geo-lift-experiment | Product page Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | The connection guide covers Claude and ChatGPT, regional server URLs, WorkOS invitations and authentication. ChatGPT navigation is less detailed, but a connection and consent workflow is documented for both clients. | https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 1 | The MCP guides explicitly describe Sellforte-rendered tables and charts inside supported clients. Claude is a documented compatible client, so this goes beyond client-generated visualizations alone. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | Sellforte documents discoverable MCP tools and runtime discovery of datasets, metrics and dimensions. The reporting guide also defines the returned metrics and explains data-model-dependent availability. | https://sellforte.com/mcp https://sellforte.com/support/marketing-reporting-with-sellforte-mcp | Product page Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The MCP guides state that results include underlying structured rows. These rows can be reused in spreadsheets, reports and subsequent calculations. | https://sellforte.com/support/marketing-reporting-with-sellforte-mcp https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Technical doc Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 0.5 | Sellforte documents agent integration and identifies Claude Code as an external runtime. I found no repeatable scheduled-invocation example or recurring-workflow authentication guidance, so automation evidence is partial. | https://sellforte.com/blog/sellforte-mcp https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client | Marketing collateral Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 1 | Sellforte documents a Bayesian MMM backbone and MCP calls into its measurement and optimization capabilities. The numeric outputs come from the model, although the surrounding LLM narrative is not established as deterministic. | https://sellforte.com/marketing-mix-modeling https://sellforte.com/support/budget-optimization-scenario-planning-with-sellforte-mcp-channel-level | Product page Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | Technical documentation describes informative MMM priors grounded in incrementality experiments. The Meta Conversion Lift workflow explains how study evidence updates priors used in daily modeling. | https://sellforte.com/support/informative-vs-non-informative-roi-priors-in-marketing-mix-modeling https://sellforte.com/blog/mmm-calibration-meta-conversion-lift-studies | Technical doc Marketing collateral |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | The platform’s Model Validation view reports R², MAPE, bias and maximum absolute percentage error. This earns full credit under the platform-level scoring rule for categories 6–8, without assuming MCP exposure. | https://sellforte.com/support/understanding-model-validation | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 1 | Enterprise pricing explicitly includes self-serve model calibration tools. The calibration documentation shows inspectable experiment-based and final priors with supporting study counts, together supporting user control and auditability at platform level. | https://sellforte.com/pricing https://sellforte.com/blog/mmm-calibration-meta-conversion-lift-studies | Product page Marketing collateral |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | I found no documented agent-controlled calibration workflow with validation and version history. Self-serve calibration and automatically refreshed experiment priors do not establish controlled calibration changes through MCP or API. | https://sellforte.com/pricing https://sellforte.com/blog/mmm-calibration-meta-conversion-lift-studies https://sellforte.com/mcp | Product page Marketing collateral Product page |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0.5 | Sellforte publicly names numerous enterprise customers and claims more than 20 customers with annual revenue above $1 billion. I could not substantiate ten individually named references with that revenue threshold using the permitted vendor sources, so evidence is partial. | https://sellforte.com/ | Product page |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | Sellforte states that NIXU, part of DNV, regularly audits its software and processes. This satisfies the criterion’s independent cybersecurity-audit alternative without implying SOC 2 or ISO 27001 certification. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0.5 | Technical documentation confirms EU default storage, alternative regions by request and separate EU and US MCP endpoints. I found no explicit commitment that customers can choose US data storage, so endpoint geography provides only partial residency evidence. | https://sellforte.com/support/where-is-my-data-located-if-i-connect-it-to-sellforte https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client | Technical doc Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 1 | Sellforte’s security page specifies AWS as the primary cloud and Azure or Google Cloud as paid alternatives. This explicitly documents all three hosting options in the criterion. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | Sellforte documents customer SSO through Azure AD and Google. The security page identifies this as an enterprise authentication option with separate pricing. | https://sellforte.com/security | Product page |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | I found no publicly accessible, promptable MCP demo or trial without access gating. The connection guide requires an individual invitation through a Sellforte contact or waitlist, while a general product tour does not demonstrate ungated MCP access. | https://sellforte.com/support/connecting-sellforte-mcp-to-your-llm-client https://sellforte.com/mcp | Technical doc Product page |
Triple Whale
Claude evaluation
Opus 5.5
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The MCP docs list revenue, orders, spend and ROAS as example questions answered from live Triple Whale data for a given day or period. The SQL API adds date-bounded queries on the blended stats and orders tables for online store sales. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.75 | Offline, retail, wholesale and POS revenue can be uploaded by location, state and ZIP into the External Revenue table, which the SQL API can query by date. We found no evidence of a direct feed of physical-store sales from the customer's own data warehouse; loading is by upload. | https://triplewhale.readme.io/docs/external-revenue-table | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The Ads table, queryable through the SQL API, holds spend, impressions, clicks and platform-reported conversions per campaign, ad set and ad. The data dictionary documents these metrics for Meta, Google, TikTok, Microsoft, Pinterest, Snapchat and other paid platforms. | https://triplewhale.readme.io/docs/ads-table | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.5 | The MMM setup wizard accepts custom media uploads of weekly spend and impressions for sources such as TV, podcast, OOH and direct mail, but this is an app workflow. We found no evidence that the MCP or API reports offline media spend or metrics, and print is not named. | https://kb.triplewhale.com/en/articles/16786415-mmm-custom-media | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The data dictionary defines queryable metrics such as Net Profit, Net Margin, orders, new customer orders and new customer revenue. The MCP docs give new-customer revenue by product as an example question. | https://triplewhale.readme.io/docs/net-profit | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 0.75 | The SQL API takes a shop ID and an explicit start and end date, and queries can group or filter by any column, such as campaign, ad set, product or channel. Brand is handled one shop per request or connection, and we found no documented market filter for the MCP. | https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The MMM Attributions table, queryable through the SQL API, returns attributed KPI and spend per modeled group, such as Meta Prospecting or Google PMAX, with 85% HDI bounds. Dividing the incremental KPI by spend gives incremental ROAS per digital channel, deduplicated by the Bayesian MMM. | https://triplewhale.readme.io/docs/mmm-attributions-table | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.5 | The MMM covers offline channels such as TV, radio and out-of-home, and custom media uploads bring them into the model. We found no evidence that the MCP or API returns incremental ROAS for offline channels specifically. | https://kb.triplewhale.com/en/articles/12325857-introduction-to-marketing-mix-modeling-mmm | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 1 | The MMM Attributions table includes non-paid controller groups such as Discounts, with the discount percentage input and the KPI attributed to it per date. This is queryable through the SQL API alongside paid media contributions. | https://triplewhale.readme.io/docs/mmm-attributions-table | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0.25 | Models can be trained at daily granularity, so the platform offers daily-level MMM outputs. The docs state that the MMM refreshes weekly, and we found no evidence of daily refreshed incremental ROAS. | https://kb.triplewhale.com/en/articles/12325857-introduction-to-marketing-mix-modeling-mmm | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 0.75 | Through the SQL API, the MMM Attributions table splits the prediction by modeled group, covering paid media and non-media controllers such as email and discounts, and MMM Predictions gives total predicted vs actual KPI. Baseline revenue is documented in the app's decomposition view, but we found no explicit baseline field in the API tables. | https://triplewhale.readme.io/docs/mmm-attributions-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.75 | Each MMM run produces a default optimized reallocation, and the MMM Reallocation Settings table records optimization scenarios with request sources chat, agent and api. We found no documented MCP tool or public API endpoint to trigger a new optimization, only to retrieve results. | https://triplewhale.readme.io/docs/mmm-reallocation-settings-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.75 | Simulation scenarios for a user-specified allocation return projected KPI and efficiency per modeled group in the MMM Reallocations table, queryable through the SQL API. The request source field lists chat, agent and api, but we found no documented MCP call for submitting a new plan. | https://triplewhale.readme.io/docs/mmm-reallocations-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | The MMM Marginal Lifts table returns each modeled group's marginal return on the next spend increment through the SQL API. Response curves are documented only as plots in the app; we found no API table for curve points. | https://triplewhale.readme.io/docs/mmm-marginal-lifts-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.75 | Simulation scenarios support exact, absolute or percent budget changes per group, and the resulting KPI and MER changes are stored in the MMM Reallocations table for API retrieval. We found no documented MCP call to submit a new budget change directly. | https://triplewhale.readme.io/docs/mmm-reallocations-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.75 | The MMM Reallocations table exposes max increase and max decrease guardrails per modeled group, and the app also supports locking subcategories. We found no evidence of passing planning dates or channel restrictions through the MCP. | https://triplewhale.readme.io/docs/mmm-reallocations-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 1 | The MMM Reallocations table returns initial budget, KPI and MER beside the new scenario budget, KPI and MER per group, with percent changes. Scenarios such as Current Budget, Scale Up and Scale Down are identified in the Reallocation Settings table and can be compared through the SQL API. | https://triplewhale.readme.io/docs/mmm-reallocations-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.75 | MMM runs can target Revenue, NC Revenue, Profit, NC Profit, Orders, NC Orders or Subscriptions, and optimization shifts budget toward the chosen KPI; these are exposed in the MMM Models table. We found no evidence of solving for a budget that reaches a specified outcome level through the MCP. | https://triplewhale.readme.io/docs/mmm-models-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 1 | The MMM Reallocation Settings table returns saved scenarios with IDs, names, type, requested budget and default flag. The MMM Reallocations table gives their per-group inputs and outputs, both queryable through the SQL API. | https://triplewhale.readme.io/docs/mmm-reallocation-settings-table | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The Pixel Joined table, queryable by campaign, ad set and ad, includes an iROAS metric based on exposed versus holdout groups, and GeoLift results report iROAS for tested campaigns. Incremental ROAS appears only where a lift test exists, and MMM results are at the modeled group level. | https://triplewhale.readme.io/docs/pixel-joined-table | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | The Pixel Joined table returns spend, Pixel revenue under models such as Last Click, and channel-reported conversion value by campaign and ad set, alongside the iROAS metric. The comparison depends on lift test coverage, although the GeoLift table also sets iROAS beside attribution ROAS for tested campaigns. | https://triplewhale.readme.io/docs/pixel-joined-table | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.25 | MMM marginal lift is reported per modeled group, and Scale Up/Scale Down tests measured the marginal lift of budget changes on campaigns. We found no evidence of modeled marginal returns for each campaign and ad set, and new Scale tests currently cannot be created. | https://kb.triplewhale.com/en/articles/16786050-scale-up-scale-down-testing-measuring-the-impact-of-budget-changes | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.5 | The MCP is documented as read-only analysis, and we found no evidence of budget recommendations through it. In the app, Moby Actions audits Meta and Google campaigns and queues recommended budget increases or decreases for approval. | https://www.triplewhale.com/blog/moby-actions-media-buying | Marketing collateral |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0 | We found no evidence of bid or Target ROAS recommendations through the MCP or in the platform. Current bid strategy and bid amount are readable as ad set fields, and Moby Actions describes budget and pause changes. | https://triplewhale.readme.io/docs/pixel-joined-table | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | The MCP docs state it cannot change budgets, campaigns or ad accounts. Moby Actions inside Triple Whale can execute approved budget changes across Meta and Google, on the Automate plan. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0 | The MCP is read-only, and we found no evidence that Moby Actions or any other part of the platform pushes bid or Target ROAS changes. Documented actions cover budget changes, pauses and creative. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.25 | Moby Actions records every completed action in an Actions log, and the Activities table logs campaign changes. We found no evidence of a pre/post analysis of revenue and spend impact per executed change. | https://www.triplewhale.com/blog/moby-actions-media-buying | Marketing collateral |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 1 | The Ads Experiments GeoLift table, queryable through the SQL API, returns one row per experiment cell with status, dates, tested campaigns, primary metric, lift, iROAS, confidence and intervals. It also includes new vs returning splits and per-sales-platform results. | https://triplewhale.readme.io/docs/ads-experiments-geolift-table | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0 | We did not find evidence of owned-media holdout or A/B tests such as email, SMS or leaflet experiments. Documented test types are GeoLift, Meta Conversion Lift and Scale Up/Scale Down. | https://kb.triplewhale.com/en/articles/12441418-incrementality-testing-in-triple-whale | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.75 | Triple Whale can launch Meta Conversion Lift experiments, and the Pixel Joined table exposes an iROAS metric computed from exposed versus holdout groups. The docs do not tie that metric to Meta Conversion Lift by name, and results are otherwise documented in the app. | https://triplewhale.readme.io/docs/pixel-joined-table | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.5 | Compass scores confidence per channel, flags where signals disagree, and points to incrementality tests as the way to raise confidence; GeoLift records a recommended test setup ID. We found no evidence that the MCP returns test recommendations. | https://kb.triplewhale.com/en/articles/15605363-how-compass-scores-confidence-and-handles-signal-disagreement | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | The GeoLift wizard proposes geo splits and holdout share, warns on feasibility, recommends duration and shows cost, and the GeoLift table stores MDE, spend reduction and pre-test confidence for launched tests. We found no evidence that the MCP or API produces a new test design. | https://kb.triplewhale.com/en/articles/12441418-incrementality-testing-in-triple-whale | Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Triple Whale publishes step-by-step MCP connection guides for Claude and ChatGPT, using the server URL and OAuth2 scoped to read-only access. The MCP overview also documents an API key option for config-based clients. | https://kb.triplewhale.com/en/articles/15463483-how-to-connect-triple-whale-to-claude-with-the-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | We found no evidence that the MCP returns server-provided charts, tables or UI components. The MCP docs say the external AI tool presents the analysis, while Moby artifacts are produced inside Triple Whale. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 0.5 | The data dictionary documents about 50 tables and individual metrics with SQL names, formulas and examples. We did not find public documentation of the MCP tool definitions themselves. | https://triplewhale.readme.io/llms.txt | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The SQL API returns complete query results as structured JSON. Data Warehouse Export pushes tables to BigQuery, Snowflake, AWS S3 or Google Cloud Storage on a schedule. | https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 1 | The MCP docs name n8n, Gumloop, Cursor and other MCP clients, and give an API key configuration for tools without OAuth2. They describe recurring workflows, such as a weekly ROAS check, running in the external tool. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 0.5 | Triple Whale's MMM uses a Bayesian framework, and Moby may use Compass MMM context when answering. MCP answers default largely to Pixel attribution data, so we found only partial evidence that MCP answers are grounded in the MMM. | https://kb.triplewhale.com/en/articles/12325857-introduction-to-marketing-mix-modeling-mmm | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | The MMM setup wizard has an option to include connected GeoLift results as priors, and subcategory Prior ROAS can be set from experiments. Evidence Links also compares MMM efficiency with test iROAS per subcategory. | https://kb.triplewhale.com/en/articles/16785325-setting-up-your-mmm-model | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | Each MMM run reports train and test R-squared, MAPE, SMAPE, WMAPE, CRPS and Pearson correlation, with a held-out test period. The same metrics are exposed in the MMM Models table through the SQL API. | https://triplewhale.readme.io/docs/mmm-models-table | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 1 | Users can edit Prior ROAS per subcategory, fill priors from Pixel attribution, and adjust advanced settings such as saturation, adstock lag, seasonality and time-varying effects in the MMM wizard. Models can be viewed, duplicated and updated from the Models table in the app. | https://kb.triplewhale.com/en/articles/16785325-setting-up-your-mmm-model | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | We found no evidence of proposing or applying calibration changes through the MCP. The MCP is documented as read-only and cannot change anything in the Triple Whale account. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0 | We did not find at least 10 publicly named customers with USD 1B+ revenue. Named case studies and logos are DTC brands such as True Classic, Origin and Travelpro, and the site gives no customer revenue figures. | https://www.triplewhale.com/case-studies | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | The Trust Center lists SOC 2 Type 2, SOC 2 Type 1 and SOC 3 reports, with annual penetration testing. We did not find ISO 27001. | https://trust.triplewhale.com | Product page |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | We found no evidence of a choice between US and EU data residency. The Trust Center lists US-based subprocessors, including Google Cloud, and covers EU transfers with the Data Privacy Framework and SCCs. | https://trust.triplewhale.com | Product page |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | We found no evidence of a choice between AWS, GCP and Azure for hosting. The Trust Center lists Google Cloud as the infrastructure provider; AWS S3 appears only as an export destination. | https://trust.triplewhale.com | Product page |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 1 | The Trust Center FAQ states that SAML is available and can be set up during onboarding, and the pricing FAQ says Enterprise customers get SSO. The Enterprise plan card still lists SSO as coming soon, and SCIM is not available. | https://trust.triplewhale.com | Product page |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | We found no public, promptable MCP demo or trial; the pricing page excludes the MCP from the free plan. Paid plans that include the MCP start with a booked walkthrough. | https://www.triplewhale.com/pricing | Product page |
ChatGPT evaluation
GPT-6 Astra
| Category ID | Category | Criteria ID | Criteria | Criteria Score | Rationale | URL to source | Type of source |
|---|---|---|---|---|---|---|---|
| 1 | Marketing Data Reporting via MCP | 1.1 | MCP reports sales progress for online sales | 1 | The MCP connection guide explicitly supports period-specific revenue and order reporting from live commerce data. The Summary Page API also accepts a start and end date, providing strong evidence of online sales retrieval. | https://kb.triplewhale.com/en/articles/15463483-how-to-connect-triple-whale-to-claude-with-the-mcp https://triplewhale.readme.io/reference/get-summary-page-data | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.2 | MCP reports sales progress for offline store sales | 0.75 | The SQL-accessible External Revenue table includes dated retail, point-of-sale and offline revenue with store and region fields. I found partial API evidence because the documented ingestion is uploaded records, rather than direct retrieval from the customer’s data warehouse. | https://triplewhale.readme.io/docs/external-revenue-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.3 | MCP reports digital media data (spend, media metrics) | 1 | The Ads table provides spend, impressions, clicks and performance by connected advertising platform, campaign and ad. The documented SQL API makes these cross-platform records available programmatically, including Meta, Google and TikTok data. | https://triplewhale.readme.io/docs/ads-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.4 | MCP reports offline media data (spend, media metrics) | 0.75 | The SQL-accessible Custom Spend table exposes dated advertising costs by channel, and custom MMM uploads accept offline spend or impressions. This is partial API evidence because I found no complete reporting schema covering both spend and media metrics for TV, out-of-home, radio and print. | https://triplewhale.readme.io/docs/custom-spend-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query https://kb.triplewhale.com/en/articles/16786415-mmm-custom-media | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.5 | MCP Reports business outcomes beyond revenue | 1 | The MCP guide explicitly demonstrates reporting orders and new-customer revenue alongside sales. These supported business outcomes satisfy the criterion beyond revenue-only reporting. | https://kb.triplewhale.com/en/articles/15707078-how-to-connect-triple-whale-to-chatgpt-with-the-mcp | Technical doc |
| 1 | Marketing Data Reporting via MCP | 1.6 | Filters and groups by business dimensions | 1 | The SQL API accepts custom queries against documented business, campaign, product and geographic dimensions with date filtering. These schemas support grouping at their stated grain, although available dimensions vary by source and brand scope depends on connected shops or configured categories. | https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query https://triplewhale.readme.io/docs/triple-whale-data-ontology https://triplewhale.readme.io/docs/custom-categories-table | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.1 | MCP reports incremental ROAS and incremental revenue for each digital channel | 1 | The MMM Attributions schema exposes modeled contribution and spend by group and date, with model metadata identifying revenue KPIs. The SQL API can retrieve these records and calculate incremental ROAS for modeled digital-channel groups, using aggregate platform rows to avoid double-counting. | https://triplewhale.readme.io/docs/mmm-attributions-table https://triplewhale.readme.io/docs/mmm-models-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.2 | MCP reports incremental ROAS and incremental revenue for each offline channel | 0.75 | Custom media documentation supports offline channels in MMM, while the attribution schema exposes modeled group contribution and spend through SQL. I found partial API evidence because specific offline-channel output examples and complete channel coverage are not documented. | https://kb.triplewhale.com/en/articles/16786415-mmm-custom-media https://triplewhale.readme.io/docs/mmm-attributions-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.3 | MCP reports promotion-driven revenue, in addition to media-driven | 1 | The MMM Attributions table documents a Discounts group, controller inputs and the KPI attributed to each modeled group. SQL retrieval therefore supports modeled promotional contribution when discounts are included in the model. | https://triplewhale.readme.io/docs/mmm-attributions-table https://kb.triplewhale.com/en/articles/15964506-setting-up-your-mmm-model https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.4 | MCP's incremental ROAS measurement is updated daily (not weekly or monthly), based on MMM | 0 | The technical MMM introduction states that models refresh weekly. I found no evidence of daily MMM-based incremental ROAS updates, and daily output granularity does not establish daily retraining. | https://kb.triplewhale.com/en/articles/12325857-introduction-to-marketing-mix-modeling-mmm https://triplewhale.readme.io/docs/mmm-models-table | Technical doc |
| 2 | Historical Performance Insights & Causal Explanation via MCP | 2.5 | Decomposes performance drivers (Base, media, non-media drivers) | 1 | MMM documentation describes baseline and media decomposition, and its attribution schema includes non-paid controller contributions such as discounts and email. These model outputs are queryable through the SQL API, supporting decomposition of the modeled performance drivers. | https://triplewhale.readme.io/docs/mmm-attributions-table https://kb.triplewhale.com/en/articles/15634447-reading-your-mmm-results https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.1 | MCP creates an optimized channel allocation | 0.75 | Saved scenario metadata distinguishes optimization from simulation and explicitly lists API as a request source. This is partial API evidence for invoking the optimizer because I found no public creation endpoint or full objective-and-period request contract. | https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://kb.triplewhale.com/en/articles/12325550-enhanced-marketing-mix-modeling-mmm | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.2 | MCP forecasts a specified media plan | 0.75 | Scenario metadata records simulations of user-specified allocations and API-origin requests, while scenario results expose projected KPI. I found partial API evidence because the public documentation does not specify the invocation endpoint for forecasting a newly submitted plan. | https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://triplewhale.readme.io/docs/mmm-reallocations-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.3 | MCP reports marginal returns and response curves | 0.75 | The SQL-accessible Marginal Lifts table returns each modeled group’s marginal return for a model run. Response curves are documented in the platform, but I found no API schema for complete curve samples or parameters. | https://triplewhale.readme.io/docs/mmm-marginal-lifts-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query https://kb.triplewhale.com/en/articles/15634447-reading-your-mmm-results | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.4 | MCP simulates a user-specified budget change | 0.75 | Scenario schemas store exact, absolute and percentage budget changes together with projected outcomes and an API request-source value. This partially evidences API-driven simulations, but I found no public request contract for submitting and calculating a new change. | https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://triplewhale.readme.io/docs/mmm-reallocations-table | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.5 | MCP enforces planning constraints | 0.75 | Scenario results expose upper and lower budget bounds, and the platform optimizer supports locking groups and limiting budget changes. This is partial API evidence because reading saved constraints and identifying API-origin scenarios do not document passing all constraints and planning dates into the engine. | https://triplewhale.readme.io/docs/mmm-reallocations-table https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://kb.triplewhale.com/en/articles/12325550-enhanced-marketing-mix-modeling-mmm | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.6 | MCP compares scenarios with a baseline plan | 1 | The Reallocations schema returns initial and new budgets, projected KPI and efficiency for each modeled group. Joining saved scenario identifiers through the SQL API supports comparison of alternatives against their explicit initial plan. | https://triplewhale.readme.io/docs/mmm-reallocations-table https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.7 | MCP plans against a business target | 0.75 | Model metadata includes profit, orders and new-customer KPIs, and saved optimization scenarios record API-origin requests. This provides partial API evidence for objective-based planning, but I found no public invocation contract or documented solver for reaching a specified outcome level. | https://triplewhale.readme.io/docs/mmm-models-table https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://kb.triplewhale.com/en/articles/12325550-enhanced-marketing-mix-modeling-mmm | Technical doc |
| 3 | Channel-Level Optimization with MCP | 3.8 | MCP retrieves saved plans and assumptions | 1 | Reallocation Settings exposes scenario identifiers, names, requested budget changes and initial budgets. Joining Reallocations through the SQL API retrieves the saved per-group outputs and constraints with those assumptions. | https://triplewhale.readme.io/docs/mmm-reallocation-settings-table https://triplewhale.readme.io/docs/mmm-reallocations-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.1 | MCP reports incremental ROAS for each campaign and ad set | 0.75 | The GeoLift table exposes measured incremental revenue and iROAS with the campaigns included in each experiment cell. This is partial API evidence because an experiment may combine campaigns, and I found no separate incremental estimate for every campaign and ad set. | https://triplewhale.readme.io/docs/ads-experiments-geolift-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.2 | For each campaign and ad set, MCP compares incremental ROAS to last-click and ad platform attribution ROAS | 0.75 | GeoLift results can be compared with last-click and other attribution models, and API schemas expose experiment campaign identifiers and platform-reported returns. I found partial API evidence because experiment-level lift does not establish separate, comparable incremental results for every campaign and ad set. | https://triplewhale.readme.io/docs/ads-experiments-geolift-table https://kb.triplewhale.com/en/articles/15637689-reading-incrementality-test-results https://triplewhale.readme.io/docs/ads-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.3 | For each campaign and ad set, MCP provides marginal returns (miROAS) | 0.75 | The SQL schemas expose marginal returns by modeled group and campaign-to-group mappings. This is partial API evidence because mapping a campaign to a group does not give each campaign or ad set its own independently modeled marginal return. | https://triplewhale.readme.io/docs/mmm-marginal-lifts-table https://triplewhale.readme.io/docs/mmm-campaign-groups-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.4 | For each campaign and ad set, MCP recommends optimal daily spend/budget | 0.25 | Moby Automations documents recurring proposals to increase or decrease campaign and ad-set daily budgets using performance rules. I found only partial platform evidence for optimal budget recommendations and no documented MCP or public API recommendation endpoint at both levels. | https://kb.triplewhale.com/en/articles/15465054-let-an-automation-manage-your-ads-with-rules | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.5 | For each campaign and ad set, MCP recommends optimal bid value (e.g., Target ROAS) | 0.25 | Moby can prepare paid-media changes and supports bid amounts, bid strategies and ROAS targets. I found only partial platform evidence for choosing optimal bid values, with no documented MCP or public API that returns those recommendations for each campaign and ad set. | https://kb.triplewhale.com/en/articles/11932150-what-are-moby-actions https://kb.triplewhale.com/en/articles/15458840-moby-actions-library-supported-actions-by-platform | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.6 | MCP can push daily spend/budget changes to Meta, Google etc. APIs | 0.5 | The Actions library explicitly supports changing Meta campaign and ad-set daily budgets and Google campaign budgets. This earns platform credit because the MCP is read-only and I found no public Triple Whale API endpoint for externally executing those changes. | https://kb.triplewhale.com/en/articles/15458840-moby-actions-library-supported-actions-by-platform https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.7 | MCP can push bidding changes (e.g., Target ROAS) to Meta, Google etc. APIs | 0.5 | The Actions library documents Meta bid and strategy changes plus Google CPA and ROAS target updates. This earns platform credit because the MCP is read-only and I found no public Triple Whale API execution endpoint for these bidding actions. | https://kb.triplewhale.com/en/articles/15458840-moby-actions-library-supported-actions-by-platform https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 4 | Campaign & Ad Set-Level Optimization with MCP | 4.8 | For each executed bidding change for a campaign or ad set, MCP provides pre/post analysis summarizing revenue and spend impact of the change | 0.75 | The SQL-accessible Activities table records bid and budget changes and documents joins to campaign, ad-set and ad performance for impact analysis. This is partial API evidence because I found no documented automatic pre/post revenue-and-spend report for every executed bidding change. | https://triplewhale.readme.io/docs/activities-table https://triplewhale.readme.io/docs/ads-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 5 | Incrementality Testing with MCP | 5.1 | MCP reports results for Geo Tests | 1 | The GeoLift schema contains experiment identifiers, intervention descriptions, campaigns, dates, completed status and incremental result fields. The SQL API can retrieve completed tests with their scope and measured outcomes. | https://triplewhale.readme.io/docs/ads-experiments-geolift-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 5 | Incrementality Testing with MCP | 5.2 | MCP reports results for Own Media A/B tests (e.g. leaflet tests) | 0 | I found no evidence of controlled owned-media incrementality results, such as email or leaflet holdout tests, in the public MCP, API or platform documentation reviewed. Email/SMS attribution and paid-media GeoLift reporting do not establish this capability. | https://kb.triplewhale.com/en/articles/12441418-incrementality-testing-in-triple-whale https://triplewhale.readme.io/docs/triple-whale-data-ontology | Technical doc |
| 5 | Incrementality Testing with MCP | 5.3 | MCP reports findings for Meta Conversion Lift tests | 0.5 | The Meta Conversion Lift guide documents creating lift experiments and interpreting their incremental results inside Triple Whale. I found no public MCP or API schema specifically exposing Meta Conversion Lift study results, so this receives platform credit. | https://kb.triplewhale.com/en/articles/10605805-meta-conversion-lift-experiment https://triplewhale.readme.io/docs/ads-experiments-geolift-table | Technical doc |
| 5 | Incrementality Testing with MCP | 5.4 | MCP prioritizes experiments | 0.25 | Compass highlights channels with weak measurement support or conflicting signals and recommends strengthening evidence through testing. This is partial platform evidence because I found no documented MCP or API prioritization workflow producing specific ranked experiment hypotheses. | https://kb.triplewhale.com/en/articles/15605363-how-compass-scores-confidence-and-handles-signal-disagreement | Technical doc |
| 5 | Incrementality Testing with MCP | 5.5 | MCP designs feasible incrementality tests | 0.5 | The platform documents matched test/control regions, pre-launch feasibility and duration assessment, while the experiment schema includes design-time minimum detectable effects and planned spend information. I found no public MCP or API workflow that generates a new complete feasible test design, so this receives platform credit. | https://kb.triplewhale.com/en/articles/15639483-how-geolift-tests-work https://triplewhale.readme.io/docs/ads-experiments-geolift-table | Technical doc |
| 6 | MCP interoperability and workflows | 6.1 | Documented instructions for conneting with Claude and ChatGPT | 1 | Separate technical guides provide the MCP server URL and OAuth connection steps for both Claude and ChatGPT. They also explain account prerequisites, read-only authorization and how to verify the connection. | https://kb.triplewhale.com/en/articles/15463483-how-to-connect-triple-whale-to-claude-with-the-mcp https://kb.triplewhale.com/en/articles/15707078-how-to-connect-triple-whale-to-chatgpt-with-the-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.2 | MCP provides native visual outputs | 0 | The MCP guide describes external AI clients creating reports and artifacts from retrieved data. I found no evidence of server-provided visual table or chart artifacts with documented native rendering in a compatible MCP client. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 6 | MCP interoperability and workflows | 6.3 | Exposes tool and metric documentation | 1 | The public developer reference documents API operations and a data dictionary with metric definitions, fields, grain and interpretation warnings. Because the evaluation rules accept APIs as equivalent to MCP, these documented interfaces and metric semantics satisfy the criterion. | https://triplewhale.readme.io/reference/introduction-to-the-triple-whale-api https://triplewhale.readme.io/docs/triple-whale-data-ontology https://triplewhale.readme.io/docs/ads-experiments-geolift-table | Technical doc |
| 6 | MCP interoperability and workflows | 6.4 | Returns reusable structured data | 1 | The custom SQL endpoint explicitly returns query results as JSON. This provides reusable structured data for downstream spreadsheets, reports and external tools. | https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query | Technical doc |
| 6 | MCP interoperability and workflows | 6.5 | Supports external agent automation | 1 | The MCP guide includes a configuration snippet, authentication instructions and support for external tools such as n8n and Cursor. It explicitly describes recurring external analysis, including scheduled ROAS checks using live Triple Whale data. | https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 7 | Analytical Backbone | 7.1 | MCP provides deterministic, model-backed answers with Bayesian MMM as backbone | 1 | Technical documentation identifies a Bayesian MMM engine, and published schemas expose stored model-run contributions, marginal returns and scenario outputs. These establish a model-backed analytical foundation, although they do not imply that every general Moby or MCP answer uses MMM. | https://kb.triplewhale.com/en/articles/12325857-introduction-to-marketing-mix-modeling-mmm https://triplewhale.readme.io/docs/mmm-models-table https://triplewhale.readme.io/docs/mmm-marginal-lifts-table | Technical doc |
| 7 | Analytical Backbone | 7.2 | The underlying MMM used by the MCP is calibrated with incrementality tests | 1 | The current setup guide includes a control to incorporate GeoLift experiments as priors and editable subcategory Prior ROAS. This establishes experimental calibration support, while separate Evidence Links can also compare uncalibrated models with independent tests. | https://kb.triplewhale.com/en/articles/15964506-setting-up-your-mmm-model https://kb.triplewhale.com/en/articles/16786641-evidence-links | Technical doc |
| 7 | Analytical Backbone | 7.3 | MCP reports model validation and other modelling KPIs | 1 | The MMM Models schema exposes training and held-out test R-squared, MAPE, CRPS and other accuracy measures. These metrics are queryable through the SQL API and also documented in the model-results interface. | https://triplewhale.readme.io/docs/mmm-models-table https://triplewhale.readme.io/reference/data-out-execute-custom-sql-query https://kb.triplewhale.com/en/articles/15634447-reading-your-mmm-results | Technical doc |
| 7 | Analytical Backbone | 7.4 | Model calibration & configuration settings (e.g., priors) are auditable and editable in a self-serve UI | 1 | The setup interface exposes Prior ROAS, saturation, adstock, likelihood and seasonality controls with a configuration review step. Model actions also include viewing settings and editing priors, providing self-service inspection and modification. | https://kb.triplewhale.com/en/articles/15964506-setting-up-your-mmm-model | Technical doc |
| 7 | Analytical Backbone | 7.5 | Supports controlled calibration through MCP | 0 | The platform supports editing priors in its UI and reading model-run information through API schemas. I found no evidence of agent-accessible calibration writes with validation and version history through MCP or a public API. | https://kb.triplewhale.com/en/articles/15964506-setting-up-your-mmm-model https://triplewhale.readme.io/docs/mmm-models-table https://kb.triplewhale.com/en/articles/15656798-triple-whale-mcp | Technical doc |
| 8 | Enterprise-Grade Platform | 8.1 | At least 10 public reference customers from $1B+ revenue brands | 0 | The company publishes customer case studies and aggregate customer-base statistics. I found no evidence on the permitted sources establishing at least ten named reference customers whose individual brand revenues exceed $1 billion. | https://www.triplewhale.com/case-studies | Marketing collateral |
| 8 | Enterprise-Grade Platform | 8.2 | SOC 2, ISO 27001, or audited IT security by a third-party cyber security auditor | 1 | The official Trust Center lists SOC 2 Type 2 compliance and related audit reports. The current plan documentation independently confirms that Triple Whale maintains SOC 2 Type 2 compliance. | https://trust.triplewhale.com/ https://kb.triplewhale.com/en/articles/16046642-understanding-triple-whale-plans-foundation-automate-and-enterprise | Technical doc |
| 8 | Enterprise-Grade Platform | 8.3 | Data residency: geography option between US and EU | 0 | I found no documented customer choice between US and EU data residency in the official security and data-processing materials reviewed. International transfer safeguards and GDPR statements do not establish selectable storage geography. | https://trust.triplewhale.com/ https://www.triplewhale.com/pages/data-processing-addendum | Technical doc |
| 8 | Enterprise-Grade Platform | 8.4 | Multi-cloud: option between AWS, GCP, and Azure | 0 | The Trust Center identifies Google Cloud Platform infrastructure. I found no evidence that customers can choose deployment across AWS, GCP and Azure, and warehouse export destinations do not demonstrate hosting choice. | https://trust.triplewhale.com/ https://triplewhale.readme.io/reference/introduction-to-data-warehouse-export | Technical doc |
| 8 | Enterprise-Grade Platform | 8.5 | Supports single sign-on (SSO) for enterprises | 0 | The technical plan guide explicitly marks seamless SSO as coming soon and says it should not be considered generally available without account confirmation. I found no stronger public evidence of currently available enterprise SSO, so roadmap language does not earn capability credit. | https://kb.triplewhale.com/en/articles/16046642-understanding-triple-whale-plans-foundation-automate-and-enterprise | Technical doc |
| 8 | Enterprise-Grade Platform | 8.6 | Hands-on demo or trial of the MCP capabilities is available without sales-call gating | 0 | The MCP connection guide requires an eligible account with a connected business, and documented access includes paid plans. I found no publicly accessible, promptable MCP demo or confirmed ungated MCP trial, despite the availability of a general free platform plan. | https://kb.triplewhale.com/en/articles/15707078-how-to-connect-triple-whale-to-chatgpt-with-the-mcp https://www.triplewhale.com/pricing | Technical doc; Product page |